[ { "50.00 percentile latency (ns)": 327821021865, "90.00 percentile latency (ns)": 589956474693, "95.00 percentile latency (ns)": 622683003765, "97.00 percentile latency (ns)": 635787363373, "99.00 percentile latency (ns)": 648944797267, "99.90 percentile latency (ns)": 654807947454, "Max latency (ns)": 655478616741, "Mean latency (ns)": 327639733235, "Min duration satisfied": "Yes", "Min latency (ns)": 185625289, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 171.173, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.854, "characteristics.samples_per_second": 171.173, "characteristics.samples_per_second.normalized_per_core": 21.396625, "characteristics.samples_per_second.normalized_per_processor": 21.396625, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9141, "ck_system": "A10x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A10x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 112200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "a154da4afd777119", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327821021865, "90.00 percentile latency (ns)": 589956474693, "95.00 percentile latency (ns)": 622683003765, "97.00 percentile latency (ns)": 635787363373, "99.00 percentile latency (ns)": 648944797267, "99.90 percentile latency (ns)": 654807947454, "Max latency (ns)": 655478616741, "Mean latency (ns)": 327639733235, "Min duration satisfied": "Yes", "Min latency (ns)": 185625289, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 171.173, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.854, "characteristics.samples_per_second": 171.173, "characteristics.samples_per_second.normalized_per_core": 21.396625, "characteristics.samples_per_second.normalized_per_processor": 21.396625, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9141, "ck_system": "A10x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A10x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 112200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "9448ab8eeea5ce1c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 307445608783, "90.00 percentile latency (ns)": 553178726350, "95.00 percentile latency (ns)": 583909307524, "97.00 percentile latency (ns)": 596200707160, "99.00 percentile latency (ns)": 608473988783, "99.90 percentile latency (ns)": 614020392584, "Max latency (ns)": 615183395824, "Mean latency (ns)": 307421658688, "Min duration satisfied": "Yes", "Min latency (ns)": 150656844, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 420.558, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (7x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 420.558, "characteristics.samples_per_second.normalized_per_core": 52.56975, "characteristics.samples_per_second.normalized_per_processor": 52.56975, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 258720, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 392, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "cab80b0e364ee5b3", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 307445608783, "90.00 percentile latency (ns)": 553178726350, "95.00 percentile latency (ns)": 583909307524, "97.00 percentile latency (ns)": 596200707160, "99.00 percentile latency (ns)": 608473988783, "99.90 percentile latency (ns)": 614020392584, "Max latency (ns)": 615183395824, "Mean latency (ns)": 307421658688, "Min duration satisfied": "Yes", "Min latency (ns)": 150656844, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 420.558, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (7x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 420.558, "characteristics.samples_per_second.normalized_per_core": 52.56975, "characteristics.samples_per_second.normalized_per_processor": 52.56975, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 258720, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 392, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "3306b544fb8c761e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 381611272928, "90.00 percentile latency (ns)": 690545310407, "95.00 percentile latency (ns)": 729426898407, "97.00 percentile latency (ns)": 744952509215, "99.00 percentile latency (ns)": 760523615028, "99.90 percentile latency (ns)": 767496755933, "Max latency (ns)": 768265019465, "Mean latency (ns)": 382827622418, "Min duration satisfied": "Yes", "Min latency (ns)": 108908822, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 188.997, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 188.997, "characteristics.samples_per_second.normalized_per_core": 47.24925, "characteristics.samples_per_second.normalized_per_processor": 47.24925, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 145200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "0669f56d40b5b2fd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 381611272928, "90.00 percentile latency (ns)": 690545310407, "95.00 percentile latency (ns)": 729426898407, "97.00 percentile latency (ns)": 744952509215, "99.00 percentile latency (ns)": 760523615028, "99.90 percentile latency (ns)": 767496755933, "Max latency (ns)": 768265019465, "Mean latency (ns)": 382827622418, "Min duration satisfied": "Yes", "Min latency (ns)": 108908822, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 188.997, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 188.997, "characteristics.samples_per_second.normalized_per_core": 47.24925, "characteristics.samples_per_second.normalized_per_processor": 47.24925, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 145200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "e6bd435b90f0c359", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 323026335977, "90.00 percentile latency (ns)": 581872035702, "95.00 percentile latency (ns)": 614191898910, "97.00 percentile latency (ns)": 627124187109, "99.00 percentile latency (ns)": 640053931817, "99.90 percentile latency (ns)": 652181806018, "Max latency (ns)": 657347905908, "Mean latency (ns)": 323129210086, "Min duration satisfied": "Yes", "Min latency (ns)": 162886070, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 224.904, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30 (4x1g.6gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 224.904, "characteristics.samples_per_second.normalized_per_core": 28.113, "characteristics.samples_per_second.normalized_per_processor": 28.113, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "A30-MIG_32x1g.6gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30-MIG_32x1g.6gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 147840, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30-MIG_32x1g.6gb_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A30-MIG-4x1g.6gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 224, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "88a2866343c0f492", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 323026335977, "90.00 percentile latency (ns)": 581872035702, "95.00 percentile latency (ns)": 614191898910, "97.00 percentile latency (ns)": 627124187109, "99.00 percentile latency (ns)": 640053931817, "99.90 percentile latency (ns)": 652181806018, "Max latency (ns)": 657347905908, "Mean latency (ns)": 323129210086, "Min duration satisfied": "Yes", "Min latency (ns)": 162886070, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 224.904, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30 (4x1g.6gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 224.904, "characteristics.samples_per_second.normalized_per_core": 28.113, "characteristics.samples_per_second.normalized_per_processor": 28.113, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "A30-MIG_32x1g.6gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30-MIG_32x1g.6gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 147840, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30-MIG_32x1g.6gb_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A30-MIG-4x1g.6gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 224, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "6dd370826f4ae48d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 305592285990, "90.00 percentile latency (ns)": 550553645292, "95.00 percentile latency (ns)": 581172703329, "97.00 percentile latency (ns)": 593424457752, "99.00 percentile latency (ns)": 605676364817, "99.90 percentile latency (ns)": 611179017975, "Max latency (ns)": 611784422921, "Mean latency (ns)": 305611441531, "Min duration satisfied": "Yes", "Min latency (ns)": 67714416, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 226.55, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 226.55, "characteristics.samples_per_second.normalized_per_core": 56.6375, "characteristics.samples_per_second.normalized_per_processor": 56.6375, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIe-80GB_aarch64x4_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 80, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Ampere Altra Q80-30", "host_processors_per_node": 1, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GB_aarch64x4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 138600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GB_aarch64x4_TRT", "system_name": "Gigabyte G242-P31 (4x A100-PCIe-80GB_aarch64, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 210, "task": "image segmentation", "task2": "image segmentation", "total_cores": 80, "uid": "7fd1540c240fa899", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 305592285990, "90.00 percentile latency (ns)": 550553645292, "95.00 percentile latency (ns)": 581172703329, "97.00 percentile latency (ns)": 593424457752, "99.00 percentile latency (ns)": 605676364817, "99.90 percentile latency (ns)": 611179017975, "Max latency (ns)": 611784422921, "Mean latency (ns)": 305611441531, "Min duration satisfied": "Yes", "Min latency (ns)": 67714416, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 226.55, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 226.55, "characteristics.samples_per_second.normalized_per_core": 56.6375, "characteristics.samples_per_second.normalized_per_processor": 56.6375, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIe-80GB_aarch64x4_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 80, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Ampere Altra Q80-30", "host_processors_per_node": 1, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GB_aarch64x4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 138600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GB_aarch64x4_TRT", "system_name": "Gigabyte G242-P31 (4x A100-PCIe-80GB_aarch64, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 210, "task": "image segmentation", "task2": "image segmentation", "total_cores": 80, "uid": "4a98a665a64cb9a2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 314816119842, "90.00 percentile latency (ns)": 566869660124, "95.00 percentile latency (ns)": 598377979506, "97.00 percentile latency (ns)": 610985362995, "99.00 percentile latency (ns)": 623594292752, "99.90 percentile latency (ns)": 629246411082, "Max latency (ns)": 629885595212, "Mean latency (ns)": 314833149744, "Min duration satisfied": "Yes", "Min latency (ns)": 127753722, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 240.996, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85407, "characteristics.samples_per_second": 240.996, "characteristics.samples_per_second.normalized_per_core": 30.1245, "characteristics.samples_per_second.normalized_per_processor": 30.1245, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.9142, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 151800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "9be2920236cac164", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 314816119842, "90.00 percentile latency (ns)": 566869660124, "95.00 percentile latency (ns)": 598377979506, "97.00 percentile latency (ns)": 610985362995, "99.00 percentile latency (ns)": 623594292752, "99.90 percentile latency (ns)": 629246411082, "Max latency (ns)": 629885595212, "Mean latency (ns)": 314833149744, "Min duration satisfied": "Yes", "Min latency (ns)": 127753722, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 240.996, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85407, "characteristics.samples_per_second": 240.996, "characteristics.samples_per_second.normalized_per_core": 30.1245, "characteristics.samples_per_second.normalized_per_processor": 30.1245, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.9142, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 151800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "383e4849a148af77", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330419283741, "90.00 percentile latency (ns)": 594945246752, "95.00 percentile latency (ns)": 628002158441, "97.00 percentile latency (ns)": 641228054368, "99.00 percentile latency (ns)": 654459808097, "99.90 percentile latency (ns)": 660401596308, "Max latency (ns)": 661076672348, "Mean latency (ns)": 330424425825, "Min duration satisfied": "Yes", "Min latency (ns)": 113570772, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 229.625, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85437, "characteristics.samples_per_second": 229.625, "characteristics.samples_per_second.normalized_per_core": 28.703125, "characteristics.samples_per_second.normalized_per_processor": 28.703125, "characteristics.tumor core": 0.8697, "characteristics.whole tumor": 0.9132, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 151800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "aaac5d01f18e480e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330419283741, "90.00 percentile latency (ns)": 594945246752, "95.00 percentile latency (ns)": 628002158441, "97.00 percentile latency (ns)": 641228054368, "99.00 percentile latency (ns)": 654459808097, "99.90 percentile latency (ns)": 660401596308, "Max latency (ns)": 661076672348, "Mean latency (ns)": 330424425825, "Min duration satisfied": "Yes", "Min latency (ns)": 113570772, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 229.625, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85437, "characteristics.samples_per_second": 229.625, "characteristics.samples_per_second.normalized_per_core": 28.703125, "characteristics.samples_per_second.normalized_per_processor": 28.703125, "characteristics.tumor core": 0.8697, "characteristics.whole tumor": 0.9132, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 151800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "aa1c0419282d3592", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 321091038124, "90.00 percentile latency (ns)": 575217037500, "95.00 percentile latency (ns)": 607240642013, "97.00 percentile latency (ns)": 620081779633, "99.00 percentile latency (ns)": 632953067699, "99.90 percentile latency (ns)": 638713824267, "Max latency (ns)": 639377571851, "Mean latency (ns)": 320279123292, "Min duration satisfied": "Yes", "Min latency (ns)": 62954875, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 425.289, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 425.289, "characteristics.samples_per_second.normalized_per_core": 53.161125, "characteristics.samples_per_second.normalized_per_processor": 53.161125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIe-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GBx8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "d49a4ac1810801cf", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 321091038124, "90.00 percentile latency (ns)": 575217037500, "95.00 percentile latency (ns)": 607240642013, "97.00 percentile latency (ns)": 620081779633, "99.00 percentile latency (ns)": 632953067699, "99.90 percentile latency (ns)": 638713824267, "Max latency (ns)": 639377571851, "Mean latency (ns)": 320279123292, "Min duration satisfied": "Yes", "Min latency (ns)": 62954875, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 425.289, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 425.289, "characteristics.samples_per_second.normalized_per_core": 53.161125, "characteristics.samples_per_second.normalized_per_processor": 53.161125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIe-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GBx8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "feba9ecf67d4b1fb", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 353118261261, "90.00 percentile latency (ns)": 636723692956, "95.00 percentile latency (ns)": 672177781781, "97.00 percentile latency (ns)": 686364572923, "99.00 percentile latency (ns)": 700515140529, "99.90 percentile latency (ns)": 706887266822, "Max latency (ns)": 707578010819, "Mean latency (ns)": 353199773038, "Min duration satisfied": "Yes", "Min latency (ns)": 97420104, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 345.121, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.power": 1936.396746817536, "characteristics.power.normalized_per_core": 242.049593352192, "characteristics.power.normalized_per_processor": 242.049593352192, "characteristics.samples_per_second": 345.121, "characteristics.samples_per_second.normalized_per_core": 43.140125, "characteristics.samples_per_second.normalized_per_processor": 43.140125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIex8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 244200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_MaxQ", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 370, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "b2485e80007acc93", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 353118261261, "90.00 percentile latency (ns)": 636723692956, "95.00 percentile latency (ns)": 672177781781, "97.00 percentile latency (ns)": 686364572923, "99.00 percentile latency (ns)": 700515140529, "99.90 percentile latency (ns)": 706887266822, "Max latency (ns)": 707578010819, "Mean latency (ns)": 353199773038, "Min duration satisfied": "Yes", "Min latency (ns)": 97420104, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 345.121, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.power": 1936.396746817536, "characteristics.power.normalized_per_core": 242.049593352192, "characteristics.power.normalized_per_processor": 242.049593352192, "characteristics.samples_per_second": 345.121, "characteristics.samples_per_second.normalized_per_core": 43.140125, "characteristics.samples_per_second.normalized_per_processor": 43.140125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIex8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 244200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_MaxQ", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 370, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "2186d4619a2ccbc7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325574080301, "90.00 percentile latency (ns)": 586046335042, "95.00 percentile latency (ns)": 618569116890, "97.00 percentile latency (ns)": 631582546241, "99.00 percentile latency (ns)": 644591687442, "99.90 percentile latency (ns)": 650442297882, "Max latency (ns)": 651089864310, "Mean latency (ns)": 325569967813, "Min duration satisfied": "Yes", "Min latency (ns)": 62821656, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 486.569, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 486.569, "characteristics.samples_per_second.normalized_per_core": 60.821125, "characteristics.samples_per_second.normalized_per_processor": 60.821125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 316800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "5e8cc9f043e3b2be", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325574080301, "90.00 percentile latency (ns)": 586046335042, "95.00 percentile latency (ns)": 618569116890, "97.00 percentile latency (ns)": 631582546241, "99.00 percentile latency (ns)": 644591687442, "99.90 percentile latency (ns)": 650442297882, "Max latency (ns)": 651089864310, "Mean latency (ns)": 325569967813, "Min duration satisfied": "Yes", "Min latency (ns)": 62821656, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 486.569, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 486.569, "characteristics.samples_per_second.normalized_per_core": 60.821125, "characteristics.samples_per_second.normalized_per_processor": 60.821125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 316800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "8b1ceb4ab23a3989", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331949780658, "90.00 percentile latency (ns)": 600471176155, "95.00 percentile latency (ns)": 634182453801, "97.00 percentile latency (ns)": 647670421717, "99.00 percentile latency (ns)": 661165483248, "99.90 percentile latency (ns)": 667247672751, "Max latency (ns)": 667911293726, "Mean latency (ns)": 332571818452, "Min duration satisfied": "Yes", "Min latency (ns)": 90090233, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 217.394, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 217.394, "characteristics.samples_per_second.normalized_per_core": 54.3485, "characteristics.samples_per_second.normalized_per_processor": 54.3485, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 145200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "21564c50aa928902", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331949780658, "90.00 percentile latency (ns)": 600471176155, "95.00 percentile latency (ns)": 634182453801, "97.00 percentile latency (ns)": 647670421717, "99.00 percentile latency (ns)": 661165483248, "99.90 percentile latency (ns)": 667247672751, "Max latency (ns)": 667911293726, "Mean latency (ns)": 332571818452, "Min duration satisfied": "Yes", "Min latency (ns)": 90090233, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 217.394, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 217.394, "characteristics.samples_per_second.normalized_per_core": 54.3485, "characteristics.samples_per_second.normalized_per_processor": 54.3485, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 145200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "69104214c3caa58e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1784929957777, "90.00 percentile latency (ns)": 3215739831709, "95.00 percentile latency (ns)": 3394670866781, "97.00 percentile latency (ns)": 3466129309037, "99.00 percentile latency (ns)": 3537723498086, "99.90 percentile latency (ns)": 3569872252614, "Max latency (ns)": 3573363428451, "Mean latency (ns)": 1785527340352, "Min duration satisfied": "Yes", "Min latency (ns)": 135117816, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6.87755, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30 (1x1g.6gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 6.87755, "characteristics.samples_per_second.normalized_per_core": 6.87755, "characteristics.samples_per_second.normalized_per_processor": 6.87755, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 24576, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 6.847, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "0a75a4714646dca8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1784929957777, "90.00 percentile latency (ns)": 3215739831709, "95.00 percentile latency (ns)": 3394670866781, "97.00 percentile latency (ns)": 3466129309037, "99.00 percentile latency (ns)": 3537723498086, "99.90 percentile latency (ns)": 3569872252614, "Max latency (ns)": 3573363428451, "Mean latency (ns)": 1785527340352, "Min duration satisfied": "Yes", "Min latency (ns)": 135117816, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6.87755, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30 (1x1g.6gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 6.87755, "characteristics.samples_per_second.normalized_per_core": 6.87755, "characteristics.samples_per_second.normalized_per_processor": 6.87755, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 24576, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 6.847, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "581226b2585b4f78", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1636002774593, "90.00 percentile latency (ns)": 2947670156111, "95.00 percentile latency (ns)": 3111476766592, "97.00 percentile latency (ns)": 3176406559873, "99.00 percentile latency (ns)": 3241449840785, "99.90 percentile latency (ns)": 3270675194517, "Max latency (ns)": 3273848456834, "Mean latency (ns)": 1636188408268, "Min duration satisfied": "Yes", "Min latency (ns)": 131408284, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7.50676, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (1x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 7.50676, "characteristics.samples_per_second.normalized_per_core": 7.50676, "characteristics.samples_per_second.normalized_per_processor": 7.50676, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 24576, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "df3c6a8dc16419d9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1636002774593, "90.00 percentile latency (ns)": 2947670156111, "95.00 percentile latency (ns)": 3111476766592, "97.00 percentile latency (ns)": 3176406559873, "99.00 percentile latency (ns)": 3241449840785, "99.90 percentile latency (ns)": 3270675194517, "Max latency (ns)": 3273848456834, "Mean latency (ns)": 1636188408268, "Min duration satisfied": "Yes", "Min latency (ns)": 131408284, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7.50676, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (1x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.8544, "characteristics.samples_per_second": 7.50676, "characteristics.samples_per_second.normalized_per_core": 7.50676, "characteristics.samples_per_second.normalized_per_processor": 7.50676, "characteristics.tumor core": 0.8698, "characteristics.whole tumor": 0.9132, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 24576, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "958ee19f1b2dd20b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 357391473904, "90.00 percentile latency (ns)": 644126261844, "95.00 percentile latency (ns)": 679976983384, "97.00 percentile latency (ns)": 694326809416, "99.00 percentile latency (ns)": 708682946058, "99.90 percentile latency (ns)": 715123841863, "Max latency (ns)": 715835390360, "Mean latency (ns)": 357567134804, "Min duration satisfied": "Yes", "Min latency (ns)": 89474044, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 202.84, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.power": 1150.9935754189937, "characteristics.power.normalized_per_core": 287.74839385474843, "characteristics.power.normalized_per_processor": 287.74839385474843, "characteristics.samples_per_second": 202.84, "characteristics.samples_per_second.normalized_per_core": 50.71, "characteristics.samples_per_second.normalized_per_processor": 50.71, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 145200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "2e54deac44a38882", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 357391473904, "90.00 percentile latency (ns)": 644126261844, "95.00 percentile latency (ns)": 679976983384, "97.00 percentile latency (ns)": 694326809416, "99.00 percentile latency (ns)": 708682946058, "99.90 percentile latency (ns)": 715123841863, "Max latency (ns)": 715835390360, "Mean latency (ns)": 357567134804, "Min duration satisfied": "Yes", "Min latency (ns)": 89474044, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 202.84, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.power": 1150.9935754189937, "characteristics.power.normalized_per_core": 287.74839385474843, "characteristics.power.normalized_per_processor": 287.74839385474843, "characteristics.samples_per_second": 202.84, "characteristics.samples_per_second.normalized_per_core": 50.71, "characteristics.samples_per_second.normalized_per_processor": 50.71, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 145200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "b4a7cbb1f014c2f3", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329116109384, "90.00 percentile latency (ns)": 593172788360, "95.00 percentile latency (ns)": 626178135818, "97.00 percentile latency (ns)": 639370016531, "99.00 percentile latency (ns)": 652574086937, "99.90 percentile latency (ns)": 658529372913, "Max latency (ns)": 659181197453, "Mean latency (ns)": 329190018765, "Min duration satisfied": "Yes", "Min latency (ns)": 103712175, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 412.512, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 412.512, "characteristics.samples_per_second.normalized_per_core": 51.564, "characteristics.samples_per_second.normalized_per_processor": 51.564, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "eb60c45d3e49d199", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329116109384, "90.00 percentile latency (ns)": 593172788360, "95.00 percentile latency (ns)": 626178135818, "97.00 percentile latency (ns)": 639370016531, "99.00 percentile latency (ns)": 652574086937, "99.90 percentile latency (ns)": 658529372913, "Max latency (ns)": 659181197453, "Mean latency (ns)": 329190018765, "Min duration satisfied": "Yes", "Min latency (ns)": 103712175, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 412.512, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 412.512, "characteristics.samples_per_second.normalized_per_core": 51.564, "characteristics.samples_per_second.normalized_per_processor": 51.564, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "e3a4ad6929f4b518", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 347150588924, "90.00 percentile latency (ns)": 625415951167, "95.00 percentile latency (ns)": 660079501516, "97.00 percentile latency (ns)": 674027187052, "99.00 percentile latency (ns)": 687865498168, "99.90 percentile latency (ns)": 694205748672, "Max latency (ns)": 694896387596, "Mean latency (ns)": 347242136996, "Min duration satisfied": "Yes", "Min latency (ns)": 82327668, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 391.31, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85437, "characteristics.samples_per_second": 391.31, "characteristics.samples_per_second.normalized_per_core": 48.91375, "characteristics.samples_per_second.normalized_per_processor": 48.91375, "characteristics.tumor core": 0.8697, "characteristics.whole tumor": 0.9132, "ck_system": "A100-PCIex8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "5acfb5dea8c823f4", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 347150588924, "90.00 percentile latency (ns)": 625415951167, "95.00 percentile latency (ns)": 660079501516, "97.00 percentile latency (ns)": 674027187052, "99.00 percentile latency (ns)": 687865498168, "99.90 percentile latency (ns)": 694205748672, "Max latency (ns)": 694896387596, "Mean latency (ns)": 347242136996, "Min duration satisfied": "Yes", "Min latency (ns)": 82327668, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 391.31, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85437, "characteristics.samples_per_second": 391.31, "characteristics.samples_per_second.normalized_per_core": 48.91375, "characteristics.samples_per_second.normalized_per_processor": 48.91375, "characteristics.tumor core": 0.8697, "characteristics.whole tumor": 0.9132, "ck_system": "A100-PCIex8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "ac4da3bdec8e922e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330911330158, "90.00 percentile latency (ns)": 595767116570, "95.00 percentile latency (ns)": 628877793277, "97.00 percentile latency (ns)": 642129140598, "99.00 percentile latency (ns)": 655369579206, "99.90 percentile latency (ns)": 661325124207, "Max latency (ns)": 661947921850, "Mean latency (ns)": 330929278910, "Min duration satisfied": "Yes", "Min latency (ns)": 151876704, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 478.587, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 478.587, "characteristics.samples_per_second.normalized_per_core": 59.823375, "characteristics.samples_per_second.normalized_per_processor": 59.823375, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 316800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "419aa18d1f3c980a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330911330158, "90.00 percentile latency (ns)": 595767116570, "95.00 percentile latency (ns)": 628877793277, "97.00 percentile latency (ns)": 642129140598, "99.00 percentile latency (ns)": 655369579206, "99.90 percentile latency (ns)": 661325124207, "Max latency (ns)": 661947921850, "Mean latency (ns)": 330929278910, "Min duration satisfied": "Yes", "Min latency (ns)": 151876704, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 478.587, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.samples_per_second": 478.587, "characteristics.samples_per_second.normalized_per_core": 59.823375, "characteristics.samples_per_second.normalized_per_processor": 59.823375, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 316800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "486a8e8cfb9ff8f5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 396608756030, "90.00 percentile latency (ns)": 714906646981, "95.00 percentile latency (ns)": 754720239657, "97.00 percentile latency (ns)": 770638963093, "99.00 percentile latency (ns)": 786555631976, "99.90 percentile latency (ns)": 793710337649, "Max latency (ns)": 794503834479, "Mean latency (ns)": 396787749903, "Min duration satisfied": "Yes", "Min latency (ns)": 61468138, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 398.739, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.power": 2960.709811320754, "characteristics.power.normalized_per_core": 370.08872641509424, "characteristics.power.normalized_per_processor": 370.08872641509424, "characteristics.samples_per_second": 398.739, "characteristics.samples_per_second.normalized_per_core": 49.842375, "characteristics.samples_per_second.normalized_per_processor": 49.842375, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 316800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "f3ebc9e2792642f7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 396608756030, "90.00 percentile latency (ns)": 714906646981, "95.00 percentile latency (ns)": 754720239657, "97.00 percentile latency (ns)": 770638963093, "99.00 percentile latency (ns)": 786555631976, "99.90 percentile latency (ns)": 793710337649, "Max latency (ns)": 794503834479, "Mean latency (ns)": 396787749903, "Min duration satisfied": "Yes", "Min latency (ns)": 61468138, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 398.739, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.85403, "characteristics.power": 2960.709811320754, "characteristics.power.normalized_per_core": 370.08872641509424, "characteristics.power.normalized_per_processor": 370.08872641509424, "characteristics.samples_per_second": 398.739, "characteristics.samples_per_second.normalized_per_core": 49.842375, "characteristics.samples_per_second.normalized_per_processor": 49.842375, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9142, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 316800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "904663a702703936", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328794432947, "90.00 percentile latency (ns)": 591867224241, "95.00 percentile latency (ns)": 624666547652, "97.00 percentile latency (ns)": 637844380988, "99.00 percentile latency (ns)": 650986947791, "99.90 percentile latency (ns)": 656936631649, "Max latency (ns)": 657562796490, "Mean latency (ns)": 328748164640, "Min duration satisfied": "Yes", "Min latency (ns)": 50227509, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 413.527, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85437, "characteristics.samples_per_second": 413.527, "characteristics.samples_per_second.normalized_per_core": 51.690875, "characteristics.samples_per_second.normalized_per_processor": 51.690875, "characteristics.tumor core": 0.8697, "characteristics.whole tumor": 0.9132, "ck_system": "A100-PCIe-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GBx8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GBx8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "4018ab529726fb3b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328794432947, "90.00 percentile latency (ns)": 591867224241, "95.00 percentile latency (ns)": 624666547652, "97.00 percentile latency (ns)": 637844380988, "99.00 percentile latency (ns)": 650986947791, "99.90 percentile latency (ns)": 656936631649, "Max latency (ns)": 657562796490, "Mean latency (ns)": 328748164640, "Min duration satisfied": "Yes", "Min latency (ns)": 50227509, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 413.527, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85437, "characteristics.samples_per_second": 413.527, "characteristics.samples_per_second.normalized_per_core": 51.690875, "characteristics.samples_per_second.normalized_per_processor": 51.690875, "characteristics.tumor core": 0.8697, "characteristics.whole tumor": 0.9132, "ck_system": "A100-PCIe-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GBx8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 271920, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GBx8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "5a99cffb2844cd2b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 323919910435, "90.00 percentile latency (ns)": 583427837442, "95.00 percentile latency (ns)": 615862379598, "97.00 percentile latency (ns)": 628826543588, "99.00 percentile latency (ns)": 641811088431, "99.90 percentile latency (ns)": 647643146718, "Max latency (ns)": 648281355573, "Mean latency (ns)": 323966658408, "Min duration satisfied": "Yes", "Min latency (ns)": 182347377, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 173.073, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.854, "characteristics.samples_per_second": 173.073, "characteristics.samples_per_second.normalized_per_core": 21.634125, "characteristics.samples_per_second.normalized_per_processor": 21.634125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9141, "ck_system": "A10x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A10x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 112200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "6e0540f1998ead57", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 323919910435, "90.00 percentile latency (ns)": 583427837442, "95.00 percentile latency (ns)": 615862379598, "97.00 percentile latency (ns)": 628826543588, "99.00 percentile latency (ns)": 641811088431, "99.90 percentile latency (ns)": 647643146718, "Max latency (ns)": 648281355573, "Mean latency (ns)": 323966658408, "Min duration satisfied": "Yes", "Min latency (ns)": 182347377, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 173.073, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7789, "characteristics.mean": 0.854, "characteristics.samples_per_second": 173.073, "characteristics.samples_per_second.normalized_per_core": 21.634125, "characteristics.samples_per_second.normalized_per_processor": 21.634125, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.9141, "ck_system": "A10x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A10x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 112200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "c45654e9a99c9927", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1222039855710, "90.00 percentile latency (ns)": 2202462074746, "95.00 percentile latency (ns)": 2324028992202, "97.00 percentile latency (ns)": 2372791190456, "99.00 percentile latency (ns)": 2422030448050, "99.90 percentile latency (ns)": 2444014732809, "Max latency (ns)": 2446234394349, "Mean latency (ns)": 1223157860552, "Min duration satisfied": "Yes", "Min latency (ns)": 1736636796, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 10.0465, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7871, "characteristics.mean": 0.85603, "characteristics.samples_per_second": 10.0465, "characteristics.samples_per_second.normalized_per_core": 0.08970089285714286, "characteristics.samples_per_second.normalized_per_processor": 2.511625, "characteristics.tumor core": 0.8679, "characteristics.whole tumor": 0.9131, "ck_system": "Triton_CPU_4S_8380Hx1", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "OpenVino 2021", "host_memory_capacity": "1536 GB", "host_memory_configuration": "12 slots / 32GB each / 3200 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 8380H CPU @ 2.70GHz", "host_processors_per_node": 4, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp32", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 112, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/Triton_CPU_4S_8380Hx1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "Tensorflow 2.4.0, OpenVino 2021, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 24576, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3928973/files/224_224_160.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "CPU Inference on Triton Inference Server", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Triton_CPU_4S_8380Hx1", "system_name": "Supermicro SYS-240P-TNRT (Cooper Lake running Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 10, "task": "image segmentation", "task2": "image segmentation", "total_cores": 112, "uid": "48dbb0a85f1ceb51", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 1222039855710, "90.00 percentile latency (ns)": 2202462074746, "95.00 percentile latency (ns)": 2324028992202, "97.00 percentile latency (ns)": 2372791190456, "99.00 percentile latency (ns)": 2422030448050, "99.90 percentile latency (ns)": 2444014732809, "Max latency (ns)": 2446234394349, "Mean latency (ns)": 1223157860552, "Min duration satisfied": "Yes", "Min latency (ns)": 1736636796, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 10.0465, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7871, "characteristics.mean": 0.85603, "characteristics.samples_per_second": 10.0465, "characteristics.samples_per_second.normalized_per_core": 0.08970089285714286, "characteristics.samples_per_second.normalized_per_processor": 2.511625, "characteristics.tumor core": 0.8679, "characteristics.whole tumor": 0.9131, "ck_system": "Triton_CPU_4S_8380Hx1", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "OpenVino 2021", "host_memory_capacity": "1536 GB", "host_memory_configuration": "12 slots / 32GB each / 3200 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 8380H CPU @ 2.70GHz", "host_processors_per_node": 4, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp32", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 112, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/Triton_CPU_4S_8380Hx1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "Tensorflow 2.4.0, OpenVino 2021, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 24576, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3928973/files/224_224_160.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "CPU Inference on Triton Inference Server", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Triton_CPU_4S_8380Hx1", "system_name": "Supermicro SYS-240P-TNRT (Cooper Lake running Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 10, "task": "image segmentation", "task2": "image segmentation", "total_cores": 112, "uid": "a82394095129b552", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 328404930329, "90.00 percentile latency (ns)": 592011657063, "95.00 percentile latency (ns)": 625115711575, "97.00 percentile latency (ns)": 638328443919, "99.00 percentile latency (ns)": 651464297522, "99.90 percentile latency (ns)": 657374216743, "Max latency (ns)": 658029784107, "Mean latency (ns)": 328640515773, "Min duration satisfied": "Yes", "Min latency (ns)": 184672563, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 170.509, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.85427, "characteristics.samples_per_second": 170.509, "characteristics.samples_per_second.normalized_per_core": 21.313625, "characteristics.samples_per_second.normalized_per_processor": 21.313625, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.914, "ck_system": "A10x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A10x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 112200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "c595f6fee8a96c1b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328404930329, "90.00 percentile latency (ns)": 592011657063, "95.00 percentile latency (ns)": 625115711575, "97.00 percentile latency (ns)": 638328443919, "99.00 percentile latency (ns)": 651464297522, "99.90 percentile latency (ns)": 657374216743, "Max latency (ns)": 658029784107, "Mean latency (ns)": 328640515773, "Min duration satisfied": "Yes", "Min latency (ns)": 184672563, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 170.509, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.85427, "characteristics.samples_per_second": 170.509, "characteristics.samples_per_second.normalized_per_core": 21.313625, "characteristics.samples_per_second.normalized_per_processor": 21.313625, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.914, "ck_system": "A10x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A10x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 112200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "49829681fe901d37", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 317177924672, "90.00 percentile latency (ns)": 570464430889, "95.00 percentile latency (ns)": 602106426516, "97.00 percentile latency (ns)": 614748161498, "99.00 percentile latency (ns)": 627440486708, "99.90 percentile latency (ns)": 633240701539, "Max latency (ns)": 634753094664, "Mean latency (ns)": 317086363412, "Min duration satisfied": "Yes", "Min latency (ns)": 156038337, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 407.592, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (7x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.78, "characteristics.mean": 0.85377, "characteristics.samples_per_second": 407.592, "characteristics.samples_per_second.normalized_per_core": 50.949, "characteristics.samples_per_second.normalized_per_processor": 50.949, "characteristics.tumor core": 0.8683, "characteristics.whole tumor": 0.913, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 258720, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 392, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "7d63ea401492fc99", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 317177924672, "90.00 percentile latency (ns)": 570464430889, "95.00 percentile latency (ns)": 602106426516, "97.00 percentile latency (ns)": 614748161498, "99.00 percentile latency (ns)": 627440486708, "99.90 percentile latency (ns)": 633240701539, "Max latency (ns)": 634753094664, "Mean latency (ns)": 317086363412, "Min duration satisfied": "Yes", "Min latency (ns)": 156038337, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 407.592, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (7x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.78, "characteristics.mean": 0.85377, "characteristics.samples_per_second": 407.592, "characteristics.samples_per_second.normalized_per_core": 50.949, "characteristics.samples_per_second.normalized_per_processor": 50.949, "characteristics.tumor core": 0.8683, "characteristics.whole tumor": 0.913, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 258720, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 392, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "0c72034a06a1eef0", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326604352541, "90.00 percentile latency (ns)": 588082192921, "95.00 percentile latency (ns)": 620754127828, "97.00 percentile latency (ns)": 633818931035, "99.00 percentile latency (ns)": 646889011798, "99.90 percentile latency (ns)": 652761177603, "Max latency (ns)": 653387119937, "Mean latency (ns)": 326646161567, "Min duration satisfied": "Yes", "Min latency (ns)": 43508463, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 60.6073, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 60.6073, "characteristics.samples_per_second.normalized_per_core": 60.6073, "characteristics.samples_per_second.normalized_per_processor": 60.6073, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 39600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "ff247999b1574577", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326604352541, "90.00 percentile latency (ns)": 588082192921, "95.00 percentile latency (ns)": 620754127828, "97.00 percentile latency (ns)": 633818931035, "99.00 percentile latency (ns)": 646889011798, "99.90 percentile latency (ns)": 652761177603, "Max latency (ns)": 653387119937, "Mean latency (ns)": 326646161567, "Min duration satisfied": "Yes", "Min latency (ns)": 43508463, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 60.6073, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 60.6073, "characteristics.samples_per_second.normalized_per_core": 60.6073, "characteristics.samples_per_second.normalized_per_processor": 60.6073, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 39600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "c6194a66961d6c2e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 2984786606684, "90.00 percentile latency (ns)": 5373047687716, "95.00 percentile latency (ns)": 5671315693608, "97.00 percentile latency (ns)": 5790838817785, "99.00 percentile latency (ns)": 5910888311286, "99.90 percentile latency (ns)": 5964550361163, "Max latency (ns)": 5970098220095, "Mean latency (ns)": 2984742424280, "Min duration satisfied": "Yes", "Min latency (ns)": 2108498321, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 4.11652, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7871, "characteristics.mean": 0.85603, "characteristics.samples_per_second": 4.11652, "characteristics.samples_per_second.normalized_per_core": 0.01715216666666667, "characteristics.samples_per_second.normalized_per_processor": 2.05826, "characteristics.tumor core": 0.8679, "characteristics.whole tumor": 0.9131, "ck_system": "Triton_CPU_2S_6258Rx1", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "OpenVino 2021.2.200", "host_memory_capacity": "768 GB", "host_memory_configuration": "6 slots / 32GB each / 2934 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp32", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 240, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/Triton_CPU_2S_6258Rx1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "OpenVino 2021.2.200, Triton 21.02; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "No", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 24576, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3928973/files/224_224_160.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "CPU Inference on Triton Inference ServerPowered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Triton_CPU_2S_6258Rx1", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 4, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "82e973692a0eb5d3", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 2984786606684, "90.00 percentile latency (ns)": 5373047687716, "95.00 percentile latency (ns)": 5671315693608, "97.00 percentile latency (ns)": 5790838817785, "99.00 percentile latency (ns)": 5910888311286, "99.90 percentile latency (ns)": 5964550361163, "Max latency (ns)": 5970098220095, "Mean latency (ns)": 2984742424280, "Min duration satisfied": "Yes", "Min latency (ns)": 2108498321, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 4.11652, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7871, "characteristics.mean": 0.85603, "characteristics.samples_per_second": 4.11652, "characteristics.samples_per_second.normalized_per_core": 0.01715216666666667, "characteristics.samples_per_second.normalized_per_processor": 2.05826, "characteristics.tumor core": 0.8679, "characteristics.whole tumor": 0.9131, "ck_system": "Triton_CPU_2S_6258Rx1", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "OpenVino 2021.2.200", "host_memory_capacity": "768 GB", "host_memory_configuration": "6 slots / 32GB each / 2934 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp32", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 240, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/Triton_CPU_2S_6258Rx1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "OpenVino 2021.2.200, Triton 21.02; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "No", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 24576, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3928973/files/224_224_160.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "CPU Inference on Triton Inference ServerPowered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Triton_CPU_2S_6258Rx1", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 4, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "3da58854688a6954", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 333695756126, "90.00 percentile latency (ns)": 600741264394, "95.00 percentile latency (ns)": 634132077328, "97.00 percentile latency (ns)": 647489040362, "99.00 percentile latency (ns)": 660847789029, "99.90 percentile latency (ns)": 666848757028, "Max latency (ns)": 667490109421, "Mean latency (ns)": 333683709980, "Min duration satisfied": "Yes", "Min latency (ns)": 61050496, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 59.3267, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 59.3267, "characteristics.samples_per_second.normalized_per_core": 59.3267, "characteristics.samples_per_second.normalized_per_processor": 59.3267, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 39600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "c89dd49818465597", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 333695756126, "90.00 percentile latency (ns)": 600741264394, "95.00 percentile latency (ns)": 634132077328, "97.00 percentile latency (ns)": 647489040362, "99.00 percentile latency (ns)": 660847789029, "99.90 percentile latency (ns)": 666848757028, "Max latency (ns)": 667490109421, "Mean latency (ns)": 333683709980, "Min duration satisfied": "Yes", "Min latency (ns)": 61050496, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 59.3267, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 59.3267, "characteristics.samples_per_second.normalized_per_core": 59.3267, "characteristics.samples_per_second.normalized_per_processor": 59.3267, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 39600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "a3f49c09a4bb288b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 319919018909, "90.00 percentile latency (ns)": 576049695705, "95.00 percentile latency (ns)": 608068256138, "97.00 percentile latency (ns)": 620883235347, "99.00 percentile latency (ns)": 633691270262, "99.90 percentile latency (ns)": 639453133233, "Max latency (ns)": 640089398325, "Mean latency (ns)": 319924070819, "Min duration satisfied": "Yes", "Min latency (ns)": 127204171, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 237.154, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 237.154, "characteristics.samples_per_second.normalized_per_core": 29.64425, "characteristics.samples_per_second.normalized_per_processor": 29.64425, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.46, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 151800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "2406b9edf180d58d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 319919018909, "90.00 percentile latency (ns)": 576049695705, "95.00 percentile latency (ns)": 608068256138, "97.00 percentile latency (ns)": 620883235347, "99.00 percentile latency (ns)": 633691270262, "99.90 percentile latency (ns)": 639453133233, "Max latency (ns)": 640089398325, "Mean latency (ns)": 319924070819, "Min duration satisfied": "Yes", "Min latency (ns)": 127204171, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 237.154, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 237.154, "characteristics.samples_per_second.normalized_per_core": 29.64425, "characteristics.samples_per_second.normalized_per_processor": 29.64425, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.46, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 151800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "c3197e6c48a68ce7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 347511072415, "90.00 percentile latency (ns)": 625764247375, "95.00 percentile latency (ns)": 660542731348, "97.00 percentile latency (ns)": 674452497545, "99.00 percentile latency (ns)": 688366357175, "99.90 percentile latency (ns)": 694620824449, "Max latency (ns)": 695303746117, "Mean latency (ns)": 347537967850, "Min duration satisfied": "Yes", "Min latency (ns)": 126364984, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 218.322, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85387, "characteristics.samples_per_second": 218.322, "characteristics.samples_per_second.normalized_per_core": 27.29025, "characteristics.samples_per_second.normalized_per_processor": 27.29025, "characteristics.tumor core": 0.8683, "characteristics.whole tumor": 0.9131, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.46, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 151800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "3768a635268840dd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 347511072415, "90.00 percentile latency (ns)": 625764247375, "95.00 percentile latency (ns)": 660542731348, "97.00 percentile latency (ns)": 674452497545, "99.00 percentile latency (ns)": 688366357175, "99.90 percentile latency (ns)": 694620824449, "Max latency (ns)": 695303746117, "Mean latency (ns)": 347537967850, "Min duration satisfied": "Yes", "Min latency (ns)": 126364984, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 218.322, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7802, "characteristics.mean": 0.85387, "characteristics.samples_per_second": 218.322, "characteristics.samples_per_second.normalized_per_core": 27.29025, "characteristics.samples_per_second.normalized_per_processor": 27.29025, "characteristics.tumor core": 0.8683, "characteristics.whole tumor": 0.9131, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.46, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 151800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT_Triton", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 230, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "252816cc54aaf84c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327864009181, "90.00 percentile latency (ns)": 590541684463, "95.00 percentile latency (ns)": 623380132539, "97.00 percentile latency (ns)": 636523868974, "99.00 percentile latency (ns)": 649650463127, "99.90 percentile latency (ns)": 655549582461, "Max latency (ns)": 656217744770, "Mean latency (ns)": 327872763723, "Min duration satisfied": "Yes", "Min latency (ns)": 97598654, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 372.133, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.power": 2260.8847792998477, "characteristics.power.normalized_per_core": 282.61059741248096, "characteristics.power.normalized_per_processor": 282.61059741248096, "characteristics.samples_per_second": 372.133, "characteristics.samples_per_second.normalized_per_core": 46.516625, "characteristics.samples_per_second.normalized_per_processor": 46.516625, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "A100-PCIex8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 244200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_MaxQ", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 370, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "47c09df9b2bcdbe5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327864009181, "90.00 percentile latency (ns)": 590541684463, "95.00 percentile latency (ns)": 623380132539, "97.00 percentile latency (ns)": 636523868974, "99.00 percentile latency (ns)": 649650463127, "99.90 percentile latency (ns)": 655549582461, "Max latency (ns)": 656217744770, "Mean latency (ns)": 327872763723, "Min duration satisfied": "Yes", "Min latency (ns)": 97598654, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 372.133, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.power": 2260.8847792998477, "characteristics.power.normalized_per_core": 282.61059741248096, "characteristics.power.normalized_per_processor": 282.61059741248096, "characteristics.samples_per_second": 372.133, "characteristics.samples_per_second.normalized_per_core": 46.516625, "characteristics.samples_per_second.normalized_per_processor": 46.516625, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "A100-PCIex8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 244200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_MaxQ", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 370, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "c54b6c98c3dd7410", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330017827368, "90.00 percentile latency (ns)": 594387546743, "95.00 percentile latency (ns)": 627493938474, "97.00 percentile latency (ns)": 640687204935, "99.00 percentile latency (ns)": 653872754633, "99.90 percentile latency (ns)": 659816340130, "Max latency (ns)": 660478380696, "Mean latency (ns)": 330037632530, "Min duration satisfied": "Yes", "Min latency (ns)": 63503811, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 479.652, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 479.652, "characteristics.samples_per_second.normalized_per_core": 59.9565, "characteristics.samples_per_second.normalized_per_processor": 59.9565, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 316800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "d28aba428b15e820", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330017827368, "90.00 percentile latency (ns)": 594387546743, "95.00 percentile latency (ns)": 627493938474, "97.00 percentile latency (ns)": 640687204935, "99.00 percentile latency (ns)": 653872754633, "99.90 percentile latency (ns)": 659816340130, "Max latency (ns)": 660478380696, "Mean latency (ns)": 330037632530, "Min duration satisfied": "Yes", "Min latency (ns)": 63503811, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 479.652, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 479.652, "characteristics.samples_per_second.normalized_per_core": 59.9565, "characteristics.samples_per_second.normalized_per_processor": 59.9565, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 316800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "efbb230dc0891ccc", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 336967693608, "90.00 percentile latency (ns)": 610010661932, "95.00 percentile latency (ns)": 644235458161, "97.00 percentile latency (ns)": 657917517929, "99.00 percentile latency (ns)": 671601132294, "99.90 percentile latency (ns)": 677755868426, "Max latency (ns)": 678438193827, "Mean latency (ns)": 337749735463, "Min duration satisfied": "Yes", "Min latency (ns)": 91350351, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 214.021, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 214.021, "characteristics.samples_per_second.normalized_per_core": 53.50525, "characteristics.samples_per_second.normalized_per_processor": 53.50525, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 145200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "406390466106b4fd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 336967693608, "90.00 percentile latency (ns)": 610010661932, "95.00 percentile latency (ns)": 644235458161, "97.00 percentile latency (ns)": 657917517929, "99.00 percentile latency (ns)": 671601132294, "99.90 percentile latency (ns)": 677755868426, "Max latency (ns)": 678438193827, "Mean latency (ns)": 337749735463, "Min duration satisfied": "Yes", "Min latency (ns)": 91350351, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 214.021, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 214.021, "characteristics.samples_per_second.normalized_per_core": 53.50525, "characteristics.samples_per_second.normalized_per_processor": 53.50525, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 145200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "f20b39a01d19323d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1708096595968, "90.00 percentile latency (ns)": 3078104979954, "95.00 percentile latency (ns)": 3249436379046, "97.00 percentile latency (ns)": 3317677035795, "99.00 percentile latency (ns)": 3386366588209, "99.90 percentile latency (ns)": 3417316307019, "Max latency (ns)": 3420678495838, "Mean latency (ns)": 1709538509271, "Min duration satisfied": "Yes", "Min latency (ns)": 137893163, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7.18454, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (1x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.78, "characteristics.mean": 0.85377, "characteristics.samples_per_second": 7.18454, "characteristics.samples_per_second.normalized_per_core": 7.18454, "characteristics.samples_per_second.normalized_per_processor": 7.18454, "characteristics.tumor core": 0.8683, "characteristics.whole tumor": 0.913, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 24576, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "e6c7bb7136b4e58a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1708096595968, "90.00 percentile latency (ns)": 3078104979954, "95.00 percentile latency (ns)": 3249436379046, "97.00 percentile latency (ns)": 3317677035795, "99.00 percentile latency (ns)": 3386366588209, "99.90 percentile latency (ns)": 3417316307019, "Max latency (ns)": 3420678495838, "Mean latency (ns)": 1709538509271, "Min duration satisfied": "Yes", "Min latency (ns)": 137893163, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7.18454, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (1x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.78, "characteristics.mean": 0.85377, "characteristics.samples_per_second": 7.18454, "characteristics.samples_per_second.normalized_per_core": 7.18454, "characteristics.samples_per_second.normalized_per_processor": 7.18454, "characteristics.tumor core": 0.8683, "characteristics.whole tumor": 0.913, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp16", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 24576, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "db17489aa2d49067", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 358046084337, "90.00 percentile latency (ns)": 645645702377, "95.00 percentile latency (ns)": 681613254355, "97.00 percentile latency (ns)": 695997325760, "99.00 percentile latency (ns)": 710389958764, "99.90 percentile latency (ns)": 716860075908, "Max latency (ns)": 717576226633, "Mean latency (ns)": 358258182534, "Min duration satisfied": "Yes", "Min latency (ns)": 91845703, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 202.348, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.power": 1285.3809192200567, "characteristics.power.normalized_per_core": 321.3452298050142, "characteristics.power.normalized_per_processor": 321.3452298050142, "characteristics.samples_per_second": 202.348, "characteristics.samples_per_second.normalized_per_core": 50.587, "characteristics.samples_per_second.normalized_per_processor": 50.587, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 145200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "30b58dae09ff481d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 358046084337, "90.00 percentile latency (ns)": 645645702377, "95.00 percentile latency (ns)": 681613254355, "97.00 percentile latency (ns)": 695997325760, "99.00 percentile latency (ns)": 710389958764, "99.90 percentile latency (ns)": 716860075908, "Max latency (ns)": 717576226633, "Mean latency (ns)": 358258182534, "Min duration satisfied": "Yes", "Min latency (ns)": 91845703, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 202.348, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.power": 1285.3809192200567, "characteristics.power.normalized_per_core": 321.3452298050142, "characteristics.power.normalized_per_processor": 321.3452298050142, "characteristics.samples_per_second": 202.348, "characteristics.samples_per_second.normalized_per_core": 50.587, "characteristics.samples_per_second.normalized_per_processor": 50.587, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "10 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 145200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 220, "task": "image segmentation", "task2": "image segmentation", "total_cores": 64, "uid": "8eb60914cbbbb203", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329639026936, "90.00 percentile latency (ns)": 594247080873, "95.00 percentile latency (ns)": 627309873680, "97.00 percentile latency (ns)": 640546947504, "99.00 percentile latency (ns)": 653780260097, "99.90 percentile latency (ns)": 659746130283, "Max latency (ns)": 660401930724, "Mean latency (ns)": 329746096415, "Min duration satisfied": "Yes", "Min latency (ns)": 99621491, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 411.749, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 411.749, "characteristics.samples_per_second.normalized_per_core": 51.468625, "characteristics.samples_per_second.normalized_per_processor": 51.468625, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 271920, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "549efa320d859152", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329639026936, "90.00 percentile latency (ns)": 594247080873, "95.00 percentile latency (ns)": 627309873680, "97.00 percentile latency (ns)": 640546947504, "99.00 percentile latency (ns)": 653780260097, "99.90 percentile latency (ns)": 659746130283, "Max latency (ns)": 660401930724, "Mean latency (ns)": 329746096415, "Min duration satisfied": "Yes", "Min latency (ns)": 99621491, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 411.749, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 411.749, "characteristics.samples_per_second.normalized_per_core": 51.468625, "characteristics.samples_per_second.normalized_per_processor": 51.468625, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 271920, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 412, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "8778a4a8c83159d0", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331067989736, "90.00 percentile latency (ns)": 596236102617, "95.00 percentile latency (ns)": 629376231079, "97.00 percentile latency (ns)": 642647899584, "99.00 percentile latency (ns)": 655896680231, "99.90 percentile latency (ns)": 661868498100, "Max latency (ns)": 662493079577, "Mean latency (ns)": 331109130950, "Min duration satisfied": "Yes", "Min latency (ns)": 96558634, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 478.194, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 478.194, "characteristics.samples_per_second.normalized_per_core": 59.77425, "characteristics.samples_per_second.normalized_per_processor": 59.77425, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 316800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "4d3a054126910df0", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331067989736, "90.00 percentile latency (ns)": 596236102617, "95.00 percentile latency (ns)": 629376231079, "97.00 percentile latency (ns)": 642647899584, "99.00 percentile latency (ns)": 655896680231, "99.90 percentile latency (ns)": 661868498100, "Max latency (ns)": 662493079577, "Mean latency (ns)": 331109130950, "Min duration satisfied": "Yes", "Min latency (ns)": 96558634, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 478.194, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.samples_per_second": 478.194, "characteristics.samples_per_second.normalized_per_core": 59.77425, "characteristics.samples_per_second.normalized_per_processor": 59.77425, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 316800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_Triton", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "0d56887c2ec4309a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 365765434087, "90.00 percentile latency (ns)": 657998942255, "95.00 percentile latency (ns)": 694517736959, "97.00 percentile latency (ns)": 709118727705, "99.00 percentile latency (ns)": 723729245304, "99.90 percentile latency (ns)": 730306203767, "Max latency (ns)": 731030811551, "Mean latency (ns)": 365584203932, "Min duration satisfied": "Yes", "Min latency (ns)": 63994201, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 433.361, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.power": 3494.1846785225716, "characteristics.power.normalized_per_core": 436.77308481532145, "characteristics.power.normalized_per_processor": 436.77308481532145, "characteristics.samples_per_second": 433.361, "characteristics.samples_per_second.normalized_per_core": 54.170125, "characteristics.samples_per_second.normalized_per_processor": 54.170125, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 316800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "e9e14eb3859cc439", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 365765434087, "90.00 percentile latency (ns)": 657998942255, "95.00 percentile latency (ns)": 694517736959, "97.00 percentile latency (ns)": 709118727705, "99.00 percentile latency (ns)": 723729245304, "99.90 percentile latency (ns)": 730306203767, "Max latency (ns)": 731030811551, "Mean latency (ns)": 365584203932, "Min duration satisfied": "Yes", "Min latency (ns)": 63994201, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 433.361, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.8543, "characteristics.power": 3494.1846785225716, "characteristics.power.normalized_per_core": 436.77308481532145, "characteristics.power.normalized_per_processor": 436.77308481532145, "characteristics.samples_per_second": 433.361, "characteristics.samples_per_second.normalized_per_core": 54.170125, "characteristics.samples_per_second.normalized_per_processor": 54.170125, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.914, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 316800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 480, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "7616bc74c02967f3", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325368883747, "90.00 percentile latency (ns)": 586492347106, "95.00 percentile latency (ns)": 619161877857, "97.00 percentile latency (ns)": 632238528002, "99.00 percentile latency (ns)": 645278500888, "99.90 percentile latency (ns)": 651144986651, "Max latency (ns)": 651789534121, "Mean latency (ns)": 325487127793, "Min duration satisfied": "Yes", "Min latency (ns)": 181827014, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 172.141, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.85427, "characteristics.samples_per_second": 172.141, "characteristics.samples_per_second.normalized_per_core": 21.517625, "characteristics.samples_per_second.normalized_per_processor": 21.517625, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.914, "ck_system": "A10x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A10x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 112200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "fa3d2ea35e4c8ba7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325368883747, "90.00 percentile latency (ns)": 586492347106, "95.00 percentile latency (ns)": 619161877857, "97.00 percentile latency (ns)": 632238528002, "99.00 percentile latency (ns)": 645278500888, "99.90 percentile latency (ns)": 651144986651, "Max latency (ns)": 651789534121, "Mean latency (ns)": 325487127793, "Min duration satisfied": "Yes", "Min latency (ns)": 181827014, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 172.141, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7798, "characteristics.mean": 0.85427, "characteristics.samples_per_second": 172.141, "characteristics.samples_per_second.normalized_per_core": 21.517625, "characteristics.samples_per_second.normalized_per_processor": 21.517625, "characteristics.tumor core": 0.869, "characteristics.whole tumor": 0.914, "ck_system": "A10x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A10x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 112200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "preview", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 170, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "400bde15190aa95b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1293532365811, "90.00 percentile latency (ns)": 2327565011960, "95.00 percentile latency (ns)": 2456520573940, "97.00 percentile latency (ns)": 2508095215125, "99.00 percentile latency (ns)": 2559755481589, "99.90 percentile latency (ns)": 2582854800251, "Max latency (ns)": 2585183617366, "Mean latency (ns)": 1293692520917, "Min duration satisfied": "Yes", "Min latency (ns)": 1822642756, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 9.50648, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7871, "characteristics.mean": 0.85603, "characteristics.samples_per_second": 9.50648, "characteristics.samples_per_second.normalized_per_core": 0.08487928571428571, "characteristics.samples_per_second.normalized_per_processor": 2.37662, "characteristics.tumor core": 0.8679, "characteristics.whole tumor": 0.9131, "ck_system": "Triton_CPU_4S_8380Hx1", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "OpenVino 2021.2.200", "host_memory_capacity": "1536 GB", "host_memory_configuration": "6 slots / 32GB each / 3200 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8380H CPU @ 2.90GHz", "host_processors_per_node": 4, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "fp32", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 112, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/Triton_CPU_4S_8380Hx1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS", "other_hardware": "", "other_software_stack": "Tensorflow 2.4.0, OpenVino 2021.2.200, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "No", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 24576, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3928973/files/224_224_160.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "CPU Inference on Triton Inference Server", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Triton_CPU_4S_8380Hx1", "system_name": "Supermicro SYS-240P-TNRT (Cooper Lake running Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 9, "task": "image segmentation", "task2": "image segmentation", "total_cores": 112, "uid": "21da9188e3e7b355", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 1293532365811, "90.00 percentile latency (ns)": 2327565011960, "95.00 percentile latency (ns)": 2456520573940, "97.00 percentile latency (ns)": 2508095215125, "99.00 percentile latency (ns)": 2559755481589, "99.90 percentile latency (ns)": 2582854800251, "Max latency (ns)": 2585183617366, "Mean latency (ns)": 1293692520917, "Min duration satisfied": "Yes", "Min latency (ns)": 1822642756, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 9.50648, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.enhancing tumor": 0.7871, "characteristics.mean": 0.85603, "characteristics.samples_per_second": 9.50648, "characteristics.samples_per_second.normalized_per_core": 0.08487928571428571, "characteristics.samples_per_second.normalized_per_processor": 2.37662, "characteristics.tumor core": 0.8679, "characteristics.whole tumor": 0.9131, "ck_system": "Triton_CPU_4S_8380Hx1", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "OpenVino 2021.2.200", "host_memory_capacity": "1536 GB", "host_memory_configuration": "6 slots / 32GB each / 3200 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8380H CPU @ 2.90GHz", "host_processors_per_node": 4, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "fp32", "key.accuracy": "characteristics.mean", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 112, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/Triton_CPU_4S_8380Hx1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS", "other_hardware": "", "other_software_stack": "Tensorflow 2.4.0, OpenVino 2021.2.200, Triton 21.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 16, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "No", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 24576, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3928973/files/224_224_160.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "CPU Inference on Triton Inference Server", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Triton_CPU_4S_8380Hx1", "system_name": "Supermicro SYS-240P-TNRT (Cooper Lake running Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 9, "task": "image segmentation", "task2": "image segmentation", "total_cores": 112, "uid": "a5e28b489baeb74c", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 38293438078, "90.00 percentile latency (ns)": 68965173171, "95.00 percentile latency (ns)": 72765884043, "97.00 percentile latency (ns)": 74285308103, "99.00 percentile latency (ns)": 75822699114, "99.90 percentile latency (ns)": 76492640668, "Max latency (ns)": 76549818537, "Mean latency (ns)": 38328488504, "Min duration satisfied": "Yes", "Min latency (ns)": 113069475, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 321.046, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7786, "characteristics.mean": 0.85333, "characteristics.samples_per_second": 321.046, "characteristics.samples_per_second.normalized_per_core": 40.13075, "characteristics.samples_per_second.normalized_per_processor": 40.13075, "characteristics.tumor core": 0.8693, "characteristics.whole tumor": 0.9121, "ck_system": "DGX-A100_A100-SXM4x8_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM4x8_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 368, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "93f044b290e58e6e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 38293438078, "90.00 percentile latency (ns)": 68965173171, "95.00 percentile latency (ns)": 72765884043, "97.00 percentile latency (ns)": 74285308103, "99.00 percentile latency (ns)": 75822699114, "99.90 percentile latency (ns)": 76492640668, "Max latency (ns)": 76549818537, "Mean latency (ns)": 38328488504, "Min duration satisfied": "Yes", "Min latency (ns)": 113069475, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 321.046, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7786, "characteristics.mean": 0.85333, "characteristics.samples_per_second": 321.046, "characteristics.samples_per_second.normalized_per_core": 40.13075, "characteristics.samples_per_second.normalized_per_processor": 40.13075, "characteristics.tumor core": 0.8693, "characteristics.whole tumor": 0.9121, "ck_system": "DGX-A100_A100-SXM4x8_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM4x8_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 368, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "abbfaa32032751e6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 83798528925, "90.00 percentile latency (ns)": 151866610058, "95.00 percentile latency (ns)": 160368637467, "97.00 percentile latency (ns)": 163841514388, "99.00 percentile latency (ns)": 167228457736, "99.90 percentile latency (ns)": 168788534040, "Max latency (ns)": 168919466091, "Mean latency (ns)": 84072623793, "Min duration satisfied": "Yes", "Min latency (ns)": 486729640, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 145.489, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 145.489, "characteristics.samples_per_second.normalized_per_core": 7.27445, "characteristics.samples_per_second.normalized_per_processor": 7.27445, "characteristics.tumor core": 0.8692, "characteristics.whole tumor": 0.9125, "ck_system": "T4x20_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x20_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x20_TRT", "system_name": "Supermicro 6049GP-TRT-OTO-29 (20x T4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 150, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "a0a410198b859b03", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 83798528925, "90.00 percentile latency (ns)": 151866610058, "95.00 percentile latency (ns)": 160368637467, "97.00 percentile latency (ns)": 163841514388, "99.00 percentile latency (ns)": 167228457736, "99.90 percentile latency (ns)": 168788534040, "Max latency (ns)": 168919466091, "Mean latency (ns)": 84072623793, "Min duration satisfied": "Yes", "Min latency (ns)": 486729640, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 145.489, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 145.489, "characteristics.samples_per_second.normalized_per_core": 7.27445, "characteristics.samples_per_second.normalized_per_processor": 7.27445, "characteristics.tumor core": 0.8692, "characteristics.whole tumor": 0.9125, "ck_system": "T4x20_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x20_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x20_TRT", "system_name": "Supermicro 6049GP-TRT-OTO-29 (20x T4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 150, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "42c2efa4bb1be1c6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 162804024049, "90.00 percentile latency (ns)": 294229290330, "95.00 percentile latency (ns)": 310678506918, "97.00 percentile latency (ns)": 317253648811, "99.00 percentile latency (ns)": 323839621425, "99.90 percentile latency (ns)": 326796986182, "Max latency (ns)": 327115909998, "Mean latency (ns)": 163032536229, "Min duration satisfied": "Yes", "Min latency (ns)": 60238111, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 75.1293, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7792, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 75.1293, "characteristics.samples_per_second.normalized_per_core": 37.56465, "characteristics.samples_per_second.normalized_per_processor": 37.56465, "characteristics.tumor core": 0.8693, "characteristics.whole tumor": 0.9125, "ck_system": "A100-PCIex2_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/A100-PCIex2_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex2_TRT", "system_name": "Gigabyte G482-Z52 (2x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 82, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "7fbc6fcc9d80478d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 162804024049, "90.00 percentile latency (ns)": 294229290330, "95.00 percentile latency (ns)": 310678506918, "97.00 percentile latency (ns)": 317253648811, "99.00 percentile latency (ns)": 323839621425, "99.90 percentile latency (ns)": 326796986182, "Max latency (ns)": 327115909998, "Mean latency (ns)": 163032536229, "Min duration satisfied": "Yes", "Min latency (ns)": 60238111, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 75.1293, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7792, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 75.1293, "characteristics.samples_per_second.normalized_per_core": 37.56465, "characteristics.samples_per_second.normalized_per_processor": 37.56465, "characteristics.tumor core": 0.8693, "characteristics.whole tumor": 0.9125, "ck_system": "A100-PCIex2_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/A100-PCIex2_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex2_TRT", "system_name": "Gigabyte G482-Z52 (2x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 82, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "fcd12b73954c9890", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 37463700705, "90.00 percentile latency (ns)": 67373926086, "95.00 percentile latency (ns)": 71121088407, "97.00 percentile latency (ns)": 72609261031, "99.00 percentile latency (ns)": 74115225225, "99.90 percentile latency (ns)": 74779919567, "Max latency (ns)": 74852976050, "Mean latency (ns)": 37458332258, "Min duration satisfied": "Yes", "Min latency (ns)": 38908535, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 328.324, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.8536, "characteristics.samples_per_second": 328.324, "characteristics.samples_per_second.normalized_per_core": 41.0405, "characteristics.samples_per_second.normalized_per_processor": 41.0405, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.9124, "ck_system": "DGX-A100_A100-SXM4x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT", "system_name": "NVIDIA DGX-A100 (8x A100-SXM4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 368, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "59c5a803bb408fa9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 37463700705, "90.00 percentile latency (ns)": 67373926086, "95.00 percentile latency (ns)": 71121088407, "97.00 percentile latency (ns)": 72609261031, "99.00 percentile latency (ns)": 74115225225, "99.90 percentile latency (ns)": 74779919567, "Max latency (ns)": 74852976050, "Mean latency (ns)": 37458332258, "Min duration satisfied": "Yes", "Min latency (ns)": 38908535, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 328.324, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.8536, "characteristics.samples_per_second": 328.324, "characteristics.samples_per_second.normalized_per_core": 41.0405, "characteristics.samples_per_second.normalized_per_processor": 41.0405, "characteristics.tumor core": 0.8691, "characteristics.whole tumor": 0.9124, "ck_system": "DGX-A100_A100-SXM4x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT", "system_name": "NVIDIA DGX-A100 (8x A100-SXM4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 368, "task": "image segmentation", "task2": "image segmentation", "total_cores": 128, "uid": "b8ca74f119d23e37", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 209802774788, "90.00 percentile latency (ns)": 378869852789, "95.00 percentile latency (ns)": 400016268369, "97.00 percentile latency (ns)": 408461201350, "99.00 percentile latency (ns)": 416929119341, "99.90 percentile latency (ns)": 420689035163, "Max latency (ns)": 421121337825, "Mean latency (ns)": 209994594078, "Min duration satisfied": "Yes", "Min latency (ns)": 406002298, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 58.3585, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 58.3585, "characteristics.samples_per_second.normalized_per_core": 7.2948125, "characteristics.samples_per_second.normalized_per_processor": 7.2948125, "characteristics.tumor core": 0.8692, "characteristics.whole tumor": 0.9125, "ck_system": "T4x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x T4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "484f850466fc5b3c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 209802774788, "90.00 percentile latency (ns)": 378869852789, "95.00 percentile latency (ns)": 400016268369, "97.00 percentile latency (ns)": 408461201350, "99.00 percentile latency (ns)": 416929119341, "99.90 percentile latency (ns)": 420689035163, "Max latency (ns)": 421121337825, "Mean latency (ns)": 209994594078, "Min duration satisfied": "Yes", "Min latency (ns)": 406002298, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 58.3585, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 58.3585, "characteristics.samples_per_second.normalized_per_core": 7.2948125, "characteristics.samples_per_second.normalized_per_processor": 7.2948125, "characteristics.tumor core": 0.8692, "characteristics.whole tumor": 0.9125, "ck_system": "T4x8_TRT", "ck_used": false, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x T4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 56, "uid": "d18f47631f401e34", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 86825483692, "90.00 percentile latency (ns)": 156841969079, "95.00 percentile latency (ns)": 165644496440, "97.00 percentile latency (ns)": 169157896932, "99.00 percentile latency (ns)": 172638223832, "99.90 percentile latency (ns)": 174256568710, "Max latency (ns)": 174422083846, "Mean latency (ns)": 86982289590, "Min duration satisfied": "Yes", "Min latency (ns)": 476520835, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 140.9, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.779, "characteristics.mean": 0.85347, "characteristics.samples_per_second": 140.9, "characteristics.samples_per_second.normalized_per_core": 7.045, "characteristics.samples_per_second.normalized_per_processor": 7.045, "characteristics.tumor core": 0.8689, "characteristics.whole tumor": 0.9125, "ck_system": "T4x20_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x20_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x20_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 150, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "23143ab8d9b92a3a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 86825483692, "90.00 percentile latency (ns)": 156841969079, "95.00 percentile latency (ns)": 165644496440, "97.00 percentile latency (ns)": 169157896932, "99.00 percentile latency (ns)": 172638223832, "99.90 percentile latency (ns)": 174256568710, "Max latency (ns)": 174422083846, "Mean latency (ns)": 86982289590, "Min duration satisfied": "Yes", "Min latency (ns)": 476520835, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 140.9, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.779, "characteristics.mean": 0.85347, "characteristics.samples_per_second": 140.9, "characteristics.samples_per_second.normalized_per_core": 7.045, "characteristics.samples_per_second.normalized_per_processor": 7.045, "characteristics.tumor core": 0.8689, "characteristics.whole tumor": 0.9125, "ck_system": "T4x20_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x20_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x20_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 150, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "1ad8bac5d636c6b5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 215817682673, "90.00 percentile latency (ns)": 388714086292, "95.00 percentile latency (ns)": 410298939115, "97.00 percentile latency (ns)": 418993783550, "99.00 percentile latency (ns)": 427651517917, "99.90 percentile latency (ns)": 431537007793, "Max latency (ns)": 431911615853, "Mean latency (ns)": 215824333091, "Min duration satisfied": "Yes", "Min latency (ns)": 402748482, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 56.9005, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 56.9005, "characteristics.samples_per_second.normalized_per_core": 7.1125625, "characteristics.samples_per_second.normalized_per_processor": 7.1125625, "characteristics.tumor core": 0.8692, "characteristics.whole tumor": 0.9125, "ck_system": "T4x8_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x8_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x8_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "4cd1762f8e40a990", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 215817682673, "90.00 percentile latency (ns)": 388714086292, "95.00 percentile latency (ns)": 410298939115, "97.00 percentile latency (ns)": 418993783550, "99.00 percentile latency (ns)": 427651517917, "99.90 percentile latency (ns)": 431537007793, "Max latency (ns)": 431911615853, "Mean latency (ns)": 215824333091, "Min duration satisfied": "Yes", "Min latency (ns)": 402748482, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 56.9005, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7793, "characteristics.mean": 0.85367, "characteristics.samples_per_second": 56.9005, "characteristics.samples_per_second.normalized_per_core": 7.1125625, "characteristics.samples_per_second.normalized_per_processor": 7.1125625, "characteristics.tumor core": 0.8692, "characteristics.whole tumor": 0.9125, "ck_system": "T4x8_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x8_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x8_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 60, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "7378c3bae1dac751", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 182598662290, "90.00 percentile latency (ns)": 328619471518, "95.00 percentile latency (ns)": 346857358959, "97.00 percentile latency (ns)": 354161839729, "99.00 percentile latency (ns)": 361529718482, "99.90 percentile latency (ns)": 364830781484, "Max latency (ns)": 365188125661, "Mean latency (ns)": 182558247444, "Min duration satisfied": "Yes", "Min latency (ns)": 138964608, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 67.2968, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7787, "characteristics.mean": 0.85317, "characteristics.samples_per_second": 67.2968, "characteristics.samples_per_second.normalized_per_core": 33.6484, "characteristics.samples_per_second.normalized_per_processor": 33.6484, "characteristics.tumor core": 0.8688, "characteristics.whole tumor": 0.912, "ck_system": "A100-PCIex2_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/A100-PCIex2_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex2_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 82, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "745d8a6c879ac6bd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 182598662290, "90.00 percentile latency (ns)": 328619471518, "95.00 percentile latency (ns)": 346857358959, "97.00 percentile latency (ns)": 354161839729, "99.00 percentile latency (ns)": 361529718482, "99.90 percentile latency (ns)": 364830781484, "Max latency (ns)": 365188125661, "Mean latency (ns)": 182558247444, "Min duration satisfied": "Yes", "Min latency (ns)": 138964608, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 67.2968, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.enhancing tumor": 0.7787, "characteristics.mean": 0.85317, "characteristics.samples_per_second": 67.2968, "characteristics.samples_per_second.normalized_per_core": 33.6484, "characteristics.samples_per_second.normalized_per_processor": 33.6484, "characteristics.tumor core": 0.8688, "characteristics.whole tumor": 0.912, "ck_system": "A100-PCIex2_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "BraTS 2019", "dataset_link": "https://www.med.upenn.edu/cbica/brats2019/data.html", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mean", "dim_y_maximize": true, "division": "closed", "formal_model": "3d-unet", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "3d-unet-99.9", "input_data_types": "int8", "key.accuracy": "characteristics.mean", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/A100-PCIex2_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 16, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "224_224_160_dyanmic_bs.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex2_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 82, "task": "image segmentation", "task2": "image segmentation", "total_cores": 240, "uid": "c57be62b34a61880", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]