[ { "50.00 percentile latency (ns)": 328741755573, "90.00 percentile latency (ns)": 592475607362, "95.00 percentile latency (ns)": 625440731611, "97.00 percentile latency (ns)": 638629063833, "99.00 percentile latency (ns)": 651807039556, "99.90 percentile latency (ns)": 657765839725, "Max latency (ns)": 658394355893, "Mean latency (ns)": 328827803826, "Min duration satisfied": "Yes", "Min latency (ns)": 113174815, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 2476.02, "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.mAP": 20.113, "characteristics.samples_per_second": 2476.02, "characteristics.samples_per_second.normalized_per_core": 309.5025, "characteristics.samples_per_second.normalized_per_processor": 309.5025, "ck_system": "A10x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 1630200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 2470, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "70402e951d7770fe", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 336900555282, "90.00 percentile latency (ns)": 606487268279, "95.00 percentile latency (ns)": 640167047114, "97.00 percentile latency (ns)": 653620055564, "99.00 percentile latency (ns)": 667104480897, "99.90 percentile latency (ns)": 678053057352, "Max latency (ns)": 682029438456, "Mean latency (ns)": 336926736501, "Min duration satisfied": "Yes", "Min latency (ns)": 303075512, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 7315.81, "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.mAP": 20.114, "characteristics.samples_per_second": 7315.81, "characteristics.samples_per_second.normalized_per_core": 914.47625, "characteristics.samples_per_second.normalized_per_processor": 914.47625, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 4989600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7560, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "d4a4339e41b089e2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 372443197361, "90.00 percentile latency (ns)": 673556525056, "95.00 percentile latency (ns)": 711278899772, "97.00 percentile latency (ns)": 726392974200, "99.00 percentile latency (ns)": 741472090908, "99.90 percentile latency (ns)": 748249784682, "Max latency (ns)": 749003393030, "Mean latency (ns)": 373151017374, "Min duration satisfied": "Yes", "Min latency (ns)": 96400204, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 3383.7, "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.mAP": 20.117, "characteristics.samples_per_second": 3383.7, "characteristics.samples_per_second.normalized_per_core": 845.925, "characteristics.samples_per_second.normalized_per_processor": 845.925, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2534400, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3840, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "654d029cd43bf4fb", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 361594803099, "90.00 percentile latency (ns)": 651346513308, "95.00 percentile latency (ns)": 687570677456, "97.00 percentile latency (ns)": 702058383809, "99.00 percentile latency (ns)": 716552347357, "99.90 percentile latency (ns)": 730169280855, "Max latency (ns)": 736593845110, "Mean latency (ns)": 361659458120, "Min duration satisfied": "Yes", "Min latency (ns)": 65003427, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 3727.43, "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.mAP": 20.116, "characteristics.samples_per_second": 3727.43, "characteristics.samples_per_second.normalized_per_core": 465.92875, "characteristics.samples_per_second.normalized_per_processor": 465.92875, "ck_system": "A30-MIG_32x1g.6gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2745600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 4160, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "04ebb0b62df4a4e8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 351157111198, "90.00 percentile latency (ns)": 632534026344, "95.00 percentile latency (ns)": 667746564744, "97.00 percentile latency (ns)": 681802698794, "99.00 percentile latency (ns)": 695857161040, "99.90 percentile latency (ns)": 702216326057, "Max latency (ns)": 702867593352, "Mean latency (ns)": 351151913809, "Min duration satisfied": "Yes", "Min latency (ns)": 165837240, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3605.8, "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.mAP": 20.117, "characteristics.samples_per_second": 3605.8, "characteristics.samples_per_second.normalized_per_core": 901.45, "characteristics.samples_per_second.normalized_per_processor": 901.45, "ck_system": "A100-PCIe-80GB_aarch64x4_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2534400, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3840, "task": "object detection", "task2": "object detection", "total_cores": 80, "uid": "c0242daaa849017e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327525747427, "90.00 percentile latency (ns)": 589626947656, "95.00 percentile latency (ns)": 622385182071, "97.00 percentile latency (ns)": 635507049497, "99.00 percentile latency (ns)": 648613542516, "99.90 percentile latency (ns)": 654509867792, "Max latency (ns)": 655116840231, "Mean latency (ns)": 327507952063, "Min duration satisfied": "Yes", "Min latency (ns)": 157510911, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3788.03, "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.mAP": 20.117, "characteristics.samples_per_second": 3788.03, "characteristics.samples_per_second.normalized_per_core": 473.50375, "characteristics.samples_per_second.normalized_per_processor": 473.50375, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2481600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3760, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "bbd909af04d29055", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328189634641, "90.00 percentile latency (ns)": 590996257353, "95.00 percentile latency (ns)": 623852421264, "97.00 percentile latency (ns)": 636981424424, "99.00 percentile latency (ns)": 650110848823, "99.90 percentile latency (ns)": 656028726743, "Max latency (ns)": 656696492480, "Mean latency (ns)": 328207381619, "Min duration satisfied": "Yes", "Min latency (ns)": 76925165, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 3778.91, "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.mAP": 20.117, "characteristics.samples_per_second": 3778.91, "characteristics.samples_per_second.normalized_per_core": 472.36375, "characteristics.samples_per_second.normalized_per_processor": 472.36375, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2481600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3760, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "973184e73d34b689", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 346669951301, "90.00 percentile latency (ns)": 624630614077, "95.00 percentile latency (ns)": 659381772168, "97.00 percentile latency (ns)": 673295796066, "99.00 percentile latency (ns)": 687176345710, "99.90 percentile latency (ns)": 693432673011, "Max latency (ns)": 694058132601, "Mean latency (ns)": 346747096729, "Min duration satisfied": "Yes", "Min latency (ns)": 183881092, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7303.13, "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.mAP": 20.117, "characteristics.samples_per_second": 7303.13, "characteristics.samples_per_second.normalized_per_core": 912.89125, "characteristics.samples_per_second.normalized_per_processor": 912.89125, "ck_system": "A100-PCIe-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5068800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7680, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "bdd21c4cbabe2f07", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326148317564, "90.00 percentile latency (ns)": 587283125286, "95.00 percentile latency (ns)": 619930024237, "97.00 percentile latency (ns)": 633000002623, "99.00 percentile latency (ns)": 646080715082, "99.90 percentile latency (ns)": 651926300990, "Max latency (ns)": 652532926729, "Mean latency (ns)": 326171277144, "Min duration satisfied": "Yes", "Min latency (ns)": 193823630, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 5866.37, "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.mAP": 20.117, "characteristics.power": 2164.566921898927, "characteristics.power.normalized_per_core": 270.57086523736587, "characteristics.power.normalized_per_processor": 270.57086523736587, "characteristics.samples_per_second": 5866.37, "characteristics.samples_per_second.normalized_per_core": 733.29625, "characteristics.samples_per_second.normalized_per_processor": 733.29625, "ck_system": "A100-PCIex8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 3828000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 5800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "cb458de91e4ec50c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327846242712, "90.00 percentile latency (ns)": 590164842431, "95.00 percentile latency (ns)": 622921499724, "97.00 percentile latency (ns)": 636038461542, "99.00 percentile latency (ns)": 649179783337, "99.90 percentile latency (ns)": 655067677307, "Max latency (ns)": 655690445566, "Mean latency (ns)": 327848203210, "Min duration satisfied": "Yes", "Min latency (ns)": 155608246, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7851.27, "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.mAP": 20.117, "characteristics.samples_per_second": 7851.27, "characteristics.samples_per_second.normalized_per_core": 981.40875, "characteristics.samples_per_second.normalized_per_processor": 981.40875, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5148000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "462dce4441a9cdd6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 371509934897, "90.00 percentile latency (ns)": 671962722834, "95.00 percentile latency (ns)": 709575583142, "97.00 percentile latency (ns)": 724625314588, "99.00 percentile latency (ns)": 739697300930, "99.90 percentile latency (ns)": 746446631875, "Max latency (ns)": 747191340955, "Mean latency (ns)": 372243042554, "Min duration satisfied": "Yes", "Min latency (ns)": 178277806, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3391.9, "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.mAP": 20.117, "characteristics.samples_per_second": 3391.9, "characteristics.samples_per_second.normalized_per_core": 847.975, "characteristics.samples_per_second.normalized_per_processor": 847.975, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2534400, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3840, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "ce2f428faf70f388", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327846622906, "90.00 percentile latency (ns)": 591295528277, "95.00 percentile latency (ns)": 624211315279, "97.00 percentile latency (ns)": 637356066187, "99.00 percentile latency (ns)": 650530794724, "99.90 percentile latency (ns)": 656461405602, "Max latency (ns)": 657088284719, "Mean latency (ns)": 327770323784, "Min duration satisfied": "Yes", "Min latency (ns)": 32073247, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 115.51, "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.mAP": 20.117, "characteristics.samples_per_second": 115.51, "characteristics.samples_per_second.normalized_per_core": 115.51, "characteristics.samples_per_second.normalized_per_processor": 115.51, "ck_system": "A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 75900, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 115, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "833f3614191bd701", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 333590611706, "90.00 percentile latency (ns)": 599604020494, "95.00 percentile latency (ns)": 633101693114, "97.00 percentile latency (ns)": 646498244287, "99.00 percentile latency (ns)": 659892851815, "99.90 percentile latency (ns)": 665634255414, "Max latency (ns)": 666202816035, "Mean latency (ns)": 333339177572, "Min duration satisfied": "Yes", "Min latency (ns)": 474104960, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 133.743, "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.mAP": 20.117, "characteristics.samples_per_second": 133.743, "characteristics.samples_per_second.normalized_per_core": 133.743, "characteristics.samples_per_second.normalized_per_processor": 133.743, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 89100, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 135, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "b34aeeb915476dfd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 384029941686, "90.00 percentile latency (ns)": 692082552082, "95.00 percentile latency (ns)": 730611925644, "97.00 percentile latency (ns)": 746042499458, "99.00 percentile latency (ns)": 761457521614, "99.90 percentile latency (ns)": 768367656069, "Max latency (ns)": 769063494265, "Mean latency (ns)": 384205271406, "Min duration satisfied": "Yes", "Min latency (ns)": 183431165, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3295.44, "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.mAP": 20.117, "characteristics.power": 1245.2193758127446, "characteristics.power.normalized_per_core": 311.30484395318615, "characteristics.power.normalized_per_processor": 311.30484395318615, "characteristics.samples_per_second": 3295.44, "characteristics.samples_per_second.normalized_per_core": 823.86, "characteristics.samples_per_second.normalized_per_processor": 823.86, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2534400, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3840, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "57d9943becdd5749", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 383886299381, "90.00 percentile latency (ns)": 691515428896, "95.00 percentile latency (ns)": 729975736504, "97.00 percentile latency (ns)": 745360020519, "99.00 percentile latency (ns)": 760730685046, "99.90 percentile latency (ns)": 767644178323, "Max latency (ns)": 768352893666, "Mean latency (ns)": 383901901881, "Min duration satisfied": "Yes", "Min latency (ns)": 196252054, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6596.97, "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.mAP": 20.117, "characteristics.samples_per_second": 6596.97, "characteristics.samples_per_second.normalized_per_core": 824.62125, "characteristics.samples_per_second.normalized_per_processor": 824.62125, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5068800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7680, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "b980f73f4154ade4", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 384799662311, "90.00 percentile latency (ns)": 693281871248, "95.00 percentile latency (ns)": 731833328386, "97.00 percentile latency (ns)": 747255120886, "99.00 percentile latency (ns)": 762683850138, "99.90 percentile latency (ns)": 769616125096, "Max latency (ns)": 770390943451, "Mean latency (ns)": 384843918416, "Min duration satisfied": "Yes", "Min latency (ns)": 95371431, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 6579.52, "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.mAP": 20.117, "characteristics.samples_per_second": 6579.52, "characteristics.samples_per_second.normalized_per_core": 822.44, "characteristics.samples_per_second.normalized_per_processor": 822.44, "ck_system": "A100-PCIex8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5068800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7680, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "c4216bd34770d95f", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331360108551, "90.00 percentile latency (ns)": 596595629325, "95.00 percentile latency (ns)": 629752187224, "97.00 percentile latency (ns)": 643021242456, "99.00 percentile latency (ns)": 656270357017, "99.90 percentile latency (ns)": 662245372031, "Max latency (ns)": 662902315112, "Mean latency (ns)": 331363870868, "Min duration satisfied": "Yes", "Min latency (ns)": 111097792, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 7765.85, "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.mAP": 20.117, "characteristics.samples_per_second": 7765.85, "characteristics.samples_per_second.normalized_per_core": 970.73125, "characteristics.samples_per_second.normalized_per_processor": 970.73125, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5148000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "ac8c643b83aeed49", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 396234696043, "90.00 percentile latency (ns)": 714588336206, "95.00 percentile latency (ns)": 754423147669, "97.00 percentile latency (ns)": 770354018419, "99.00 percentile latency (ns)": 786277453351, "99.90 percentile latency (ns)": 793437982146, "Max latency (ns)": 794201430669, "Mean latency (ns)": 396482272162, "Min duration satisfied": "Yes", "Min latency (ns)": 172320800, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6481.98, "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.mAP": 20.117, "characteristics.power": 3171.4148614609558, "characteristics.power.normalized_per_core": 396.42685768261947, "characteristics.power.normalized_per_processor": 396.42685768261947, "characteristics.samples_per_second": 6481.98, "characteristics.samples_per_second.normalized_per_core": 810.2475, "characteristics.samples_per_second.normalized_per_processor": 810.2475, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5148000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "d9cc926fda3a8a6e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 347394346319, "90.00 percentile latency (ns)": 626008948444, "95.00 percentile latency (ns)": 660847093740, "97.00 percentile latency (ns)": 674767414245, "99.00 percentile latency (ns)": 688702760667, "99.90 percentile latency (ns)": 694970564218, "Max latency (ns)": 695655127553, "Mean latency (ns)": 347473749136, "Min duration satisfied": "Yes", "Min latency (ns)": 92169701, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 7286.37, "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.mAP": 20.117, "characteristics.samples_per_second": 7286.37, "characteristics.samples_per_second.normalized_per_core": 910.79625, "characteristics.samples_per_second.normalized_per_processor": 910.79625, "ck_system": "A100-PCIe-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5068800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7680, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "c24f0baaa8273878", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330497245829, "90.00 percentile latency (ns)": 595553763755, "95.00 percentile latency (ns)": 628667095562, "97.00 percentile latency (ns)": 641905423281, "99.00 percentile latency (ns)": 655159499504, "99.90 percentile latency (ns)": 661122208030, "Max latency (ns)": 661770508101, "Mean latency (ns)": 330502722203, "Min duration satisfied": "Yes", "Min latency (ns)": 128105290, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2473.36, "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.mAP": 20.113, "characteristics.samples_per_second": 2473.36, "characteristics.samples_per_second.normalized_per_core": 309.17, "characteristics.samples_per_second.normalized_per_processor": 309.17, "ck_system": "A10x8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 1636800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 2480, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "973ad4d10a0e71f5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331067839953, "90.00 percentile latency (ns)": 596078510090, "95.00 percentile latency (ns)": 629013194551, "97.00 percentile latency (ns)": 642270984669, "99.00 percentile latency (ns)": 655617877674, "99.90 percentile latency (ns)": 661480271469, "Max latency (ns)": 662053137050, "Mean latency (ns)": 331207179185, "Min duration satisfied": "Yes", "Min latency (ns)": 871929862, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 116.637, "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.mAP": 19.858, "characteristics.samples_per_second": 116.637, "characteristics.samples_per_second.normalized_per_core": 1.0414017857142857, "characteristics.samples_per_second.normalized_per_processor": 29.15925, "ck_system": "Triton_CPU_4S_8380Hx1", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "fp32", "key.accuracy": "characteristics.mAP", "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": 64, "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": 77220, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3228411/files/resnet34-ssd1200.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": 117, "task": "object detection", "task2": "object detection", "total_cores": 112, "uid": "103fd44ceead5c9c", "use_accelerator": false, "weight_data_types": "int8", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 327574410523, "90.00 percentile latency (ns)": 590307086976, "95.00 percentile latency (ns)": 623105845692, "97.00 percentile latency (ns)": 636238596901, "99.00 percentile latency (ns)": 649393477374, "99.90 percentile latency (ns)": 655276087439, "Max latency (ns)": 655947565036, "Mean latency (ns)": 327602292135, "Min duration satisfied": "Yes", "Min latency (ns)": 109954610, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 2495.32, "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.mAP": 20.111, "characteristics.samples_per_second": 2495.32, "characteristics.samples_per_second.normalized_per_core": 311.915, "characteristics.samples_per_second.normalized_per_processor": 311.915, "ck_system": "A10x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 1636800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 2480, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "3be0bb8154df4935", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 337447251427, "90.00 percentile latency (ns)": 607613821729, "95.00 percentile latency (ns)": 641402444720, "97.00 percentile latency (ns)": 654893453073, "99.00 percentile latency (ns)": 668711643163, "99.90 percentile latency (ns)": 680081682021, "Max latency (ns)": 685045578549, "Mean latency (ns)": 337527846921, "Min duration satisfied": "Yes", "Min latency (ns)": 305261702, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 7283.6, "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.mAP": 20.111, "characteristics.samples_per_second": 7283.6, "characteristics.samples_per_second.normalized_per_core": 910.45, "characteristics.samples_per_second.normalized_per_processor": 910.45, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_56x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 4989600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7560, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "8ab95a6a71d5124b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 320264565293, "90.00 percentile latency (ns)": 576562877898, "95.00 percentile latency (ns)": 608537919834, "97.00 percentile latency (ns)": 621351693650, "99.00 percentile latency (ns)": 634161279644, "99.90 percentile latency (ns)": 639973991175, "Max latency (ns)": 640491612199, "Mean latency (ns)": 320218700825, "Min duration satisfied": "Yes", "Min latency (ns)": 128797096, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 989.24, "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.mAP": 20.111, "characteristics.samples_per_second": 989.24, "characteristics.samples_per_second.normalized_per_core": 989.24, "characteristics.samples_per_second.normalized_per_processor": 989.24, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 633600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "250af29406e9e7de", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 337430866301, "90.00 percentile latency (ns)": 606567407868, "95.00 percentile latency (ns)": 640241113212, "97.00 percentile latency (ns)": 653721103294, "99.00 percentile latency (ns)": 667162053493, "99.90 percentile latency (ns)": 673130533799, "Max latency (ns)": 673693595825, "Mean latency (ns)": 337309253135, "Min duration satisfied": "Yes", "Min latency (ns)": 1033726779, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 49.9634, "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.mAP": 19.858, "characteristics.samples_per_second": 49.9634, "characteristics.samples_per_second.normalized_per_core": 0.20818083333333334, "characteristics.samples_per_second.normalized_per_processor": 24.9817, "ck_system": "Triton_CPU_2S_6258Rx1", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "fp32", "key.accuracy": "characteristics.mAP", "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": 64, "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": 33660, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3228411/files/resnet34-ssd1200.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": 51, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "2aa164615a54b2b6", "use_accelerator": false, "weight_data_types": "int8", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 321370061522, "90.00 percentile latency (ns)": 578659440729, "95.00 percentile latency (ns)": 610812579253, "97.00 percentile latency (ns)": 623679949336, "99.00 percentile latency (ns)": 636544225590, "99.90 percentile latency (ns)": 642326254172, "Max latency (ns)": 642910930576, "Mean latency (ns)": 321364843771, "Min duration satisfied": "Yes", "Min latency (ns)": 64015399, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 985.518, "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.mAP": 20.111, "characteristics.samples_per_second": 985.518, "characteristics.samples_per_second.normalized_per_core": 985.518, "characteristics.samples_per_second.normalized_per_processor": 985.518, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_Triton_datacenter", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 633600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "32c6a5ac02c5435d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330365884247, "90.00 percentile latency (ns)": 594695520159, "95.00 percentile latency (ns)": 627740585791, "97.00 percentile latency (ns)": 640952937227, "99.00 percentile latency (ns)": 654184149008, "99.90 percentile latency (ns)": 660133263752, "Max latency (ns)": 660743714015, "Mean latency (ns)": 330349222070, "Min duration satisfied": "Yes", "Min latency (ns)": 161691849, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3755.77, "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.mAP": 20.111, "characteristics.samples_per_second": 3755.77, "characteristics.samples_per_second.normalized_per_core": 469.47125, "characteristics.samples_per_second.normalized_per_processor": 469.47125, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2481600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3760, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "fd135cfd3934fa6b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329420628890, "90.00 percentile latency (ns)": 593031832251, "95.00 percentile latency (ns)": 625966837841, "97.00 percentile latency (ns)": 639146416012, "99.00 percentile latency (ns)": 652337559485, "99.90 percentile latency (ns)": 658264058058, "Max latency (ns)": 658916410232, "Mean latency (ns)": 329393789445, "Min duration satisfied": "Yes", "Min latency (ns)": 76333117, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 3766.18, "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.mAP": 20.111, "characteristics.samples_per_second": 3766.18, "characteristics.samples_per_second.normalized_per_core": 470.7725, "characteristics.samples_per_second.normalized_per_processor": 470.7725, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2481600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3760, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "920fc04013a59860", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328402716772, "90.00 percentile latency (ns)": 591554875525, "95.00 percentile latency (ns)": 624469110698, "97.00 percentile latency (ns)": 637639955294, "99.00 percentile latency (ns)": 650787917616, "99.90 percentile latency (ns)": 656700056950, "Max latency (ns)": 657307516827, "Mean latency (ns)": 328456795204, "Min duration satisfied": "Yes", "Min latency (ns)": 212814563, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 5823.76, "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.mAP": 20.111, "characteristics.power": 2244.0585996955815, "characteristics.power.normalized_per_core": 280.5073249619477, "characteristics.power.normalized_per_processor": 280.5073249619477, "characteristics.samples_per_second": 5823.76, "characteristics.samples_per_second.normalized_per_core": 727.97, "characteristics.samples_per_second.normalized_per_processor": 727.97, "ck_system": "A100-PCIex8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 3828000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 5800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "2d1336c19845c6a2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326517797686, "90.00 percentile latency (ns)": 587998552822, "95.00 percentile latency (ns)": 620666257163, "97.00 percentile latency (ns)": 633718759505, "99.00 percentile latency (ns)": 646809270727, "99.90 percentile latency (ns)": 652707950238, "Max latency (ns)": 653342446661, "Mean latency (ns)": 326556367049, "Min duration satisfied": "Yes", "Min latency (ns)": 160685227, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7879.48, "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.mAP": 20.111, "characteristics.samples_per_second": 7879.48, "characteristics.samples_per_second.normalized_per_core": 984.935, "characteristics.samples_per_second.normalized_per_processor": 984.935, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5148000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "18c048a643ceadfa", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 373818762656, "90.00 percentile latency (ns)": 676383304534, "95.00 percentile latency (ns)": 714226431634, "97.00 percentile latency (ns)": 729404369852, "99.00 percentile latency (ns)": 744544674229, "99.90 percentile latency (ns)": 751346992272, "Max latency (ns)": 752060867852, "Mean latency (ns)": 374565991714, "Min duration satisfied": "Yes", "Min latency (ns)": 186982342, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3369.94, "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.mAP": 20.111, "characteristics.samples_per_second": 3369.94, "characteristics.samples_per_second.normalized_per_core": 842.485, "characteristics.samples_per_second.normalized_per_processor": 842.485, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2534400, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3840, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "b1f44a897d612d49", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 333635239659, "90.00 percentile latency (ns)": 599642793344, "95.00 percentile latency (ns)": 633129465394, "97.00 percentile latency (ns)": 646530954155, "99.00 percentile latency (ns)": 659926658639, "99.90 percentile latency (ns)": 665672419683, "Max latency (ns)": 666241931174, "Mean latency (ns)": 333431654962, "Min duration satisfied": "Yes", "Min latency (ns)": 479184833, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 133.735, "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.mAP": 20.111, "characteristics.samples_per_second": 133.735, "characteristics.samples_per_second.normalized_per_core": 133.735, "characteristics.samples_per_second.normalized_per_processor": 133.735, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 89100, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 135, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "ef29a71744365957", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 406745722907, "90.00 percentile latency (ns)": 733281665658, "95.00 percentile latency (ns)": 774193331972, "97.00 percentile latency (ns)": 790553358341, "99.00 percentile latency (ns)": 806902681982, "99.90 percentile latency (ns)": 814246956500, "Max latency (ns)": 815012768613, "Mean latency (ns)": 406908569410, "Min duration satisfied": "Yes", "Min latency (ns)": 195567530, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3109.64, "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.mAP": 20.111, "characteristics.power": 1275.1095705521473, "characteristics.power.normalized_per_core": 318.7773926380368, "characteristics.power.normalized_per_processor": 318.7773926380368, "characteristics.samples_per_second": 3109.64, "characteristics.samples_per_second.normalized_per_core": 777.41, "characteristics.samples_per_second.normalized_per_processor": 777.41, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 2534400, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 3840, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "35c891d7594d83ad", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 384385355752, "90.00 percentile latency (ns)": 692681209242, "95.00 percentile latency (ns)": 731214446616, "97.00 percentile latency (ns)": 746616470204, "99.00 percentile latency (ns)": 762032375616, "99.90 percentile latency (ns)": 768984458436, "Max latency (ns)": 769686117664, "Mean latency (ns)": 384472883128, "Min duration satisfied": "Yes", "Min latency (ns)": 176038912, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6585.54, "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.mAP": 20.111, "characteristics.samples_per_second": 6585.54, "characteristics.samples_per_second.normalized_per_core": 823.1925, "characteristics.samples_per_second.normalized_per_processor": 823.1925, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5068800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7680, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "76a7bbb614d039aa", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327612408243, "90.00 percentile latency (ns)": 589918158944, "95.00 percentile latency (ns)": 622721431982, "97.00 percentile latency (ns)": 635840694415, "99.00 percentile latency (ns)": 648957775774, "99.90 percentile latency (ns)": 654860473882, "Max latency (ns)": 655506716770, "Mean latency (ns)": 327630884219, "Min duration satisfied": "Yes", "Min latency (ns)": 108629099, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 7853.47, "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.mAP": 20.111, "characteristics.samples_per_second": 7853.47, "characteristics.samples_per_second.normalized_per_core": 981.68375, "characteristics.samples_per_second.normalized_per_processor": 981.68375, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5148000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "17a643debc97e476", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 374466390061, "90.00 percentile latency (ns)": 673989726624, "95.00 percentile latency (ns)": 711393490200, "97.00 percentile latency (ns)": 726376141461, "99.00 percentile latency (ns)": 741350513376, "99.90 percentile latency (ns)": 748094981606, "Max latency (ns)": 748817193377, "Mean latency (ns)": 374436828157, "Min duration satisfied": "Yes", "Min latency (ns)": 172340458, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6874.84, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 20.111, "characteristics.power": 3503.1100267379707, "characteristics.power.normalized_per_core": 437.88875334224633, "characteristics.power.normalized_per_processor": 437.88875334224633, "characteristics.samples_per_second": 6874.84, "characteristics.samples_per_second.normalized_per_core": 859.355, "characteristics.samples_per_second.normalized_per_processor": 859.355, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 5148000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "f8adb1350f14922b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327679292252, "90.00 percentile latency (ns)": 590230712394, "95.00 percentile latency (ns)": 623047985756, "97.00 percentile latency (ns)": 636169646815, "99.00 percentile latency (ns)": 649298036376, "99.90 percentile latency (ns)": 655199445415, "Max latency (ns)": 655870851098, "Mean latency (ns)": 327649023828, "Min duration satisfied": "Yes", "Min latency (ns)": 124789670, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2495.61, "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.mAP": 20.111, "characteristics.samples_per_second": 2495.61, "characteristics.samples_per_second.normalized_per_core": 311.95125, "characteristics.samples_per_second.normalized_per_processor": 311.95125, "ck_system": "A10x8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "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": 1636800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 2480, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "8f214e64fd8921c2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330645868815, "90.00 percentile latency (ns)": 594675652382, "95.00 percentile latency (ns)": 627649206139, "97.00 percentile latency (ns)": 640879101592, "99.00 percentile latency (ns)": 654062066282, "99.90 percentile latency (ns)": 659993687306, "Max latency (ns)": 660477990941, "Mean latency (ns)": 330617303184, "Min duration satisfied": "Yes", "Min latency (ns)": 933172395, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 111.919, "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.mAP": 19.858, "characteristics.samples_per_second": 111.919, "characteristics.samples_per_second.normalized_per_core": 0.9992767857142857, "characteristics.samples_per_second.normalized_per_processor": 27.97975, "ck_system": "Triton_CPU_4S_8380Hx1", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "fp32", "key.accuracy": "characteristics.mAP", "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": 64, "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": 73920, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/3228411/files/resnet34-ssd1200.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": 112, "task": "object detection", "task2": "object detection", "total_cores": 112, "uid": "d934fac8ef303153", "use_accelerator": false, "weight_data_types": "int8", "weight_transformations": "We transfer the weight from fp32 datatype in ONNX file to int8 datatype in OpenVino IR file." }, { "50.00 percentile latency (ns)": 32590041184, "90.00 percentile latency (ns)": 58637430271, "95.00 percentile latency (ns)": 61892713301, "97.00 percentile latency (ns)": 63199303457, "99.00 percentile latency (ns)": 64499614462, "99.90 percentile latency (ns)": 65087471084, "Max latency (ns)": 65148797534, "Mean latency (ns)": 32593705296, "Min duration satisfied": "Yes", "Min latency (ns)": 67714790, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 91779.7, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA TITAN RTX", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 91779.7, "characteristics.samples_per_second.normalized_per_core": 22944.925, "characteristics.samples_per_second.normalized_per_processor": 22944.925, "ck_system": "TitanRTXx4", "ck_used": false, "cooling": "watercooled", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) 8268", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "ssd-small", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/TitanRTXx4", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 256, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 5979336, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "frozen_inference_graph.pb", "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/TitanRTXx4", "system_name": "SCAN 3XS DBP T496X2 Fluid", "target_latency (ns)": 0, "target_qps": 90596, "task": "object detection", "task2": "object detection", "total_cores": 48, "uid": "b60d653f97440780", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32337106921, "90.00 percentile latency (ns)": 58184487813, "95.00 percentile latency (ns)": 61437358509, "97.00 percentile latency (ns)": 62754189290, "99.00 percentile latency (ns)": 64065019932, "99.90 percentile latency (ns)": 64613028998, "Max latency (ns)": 64677492021, "Mean latency (ns)": 32346601073, "Min duration satisfied": "Yes", "Min latency (ns)": 86248046, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1661.29, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA TITAN RTX", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 20.067, "characteristics.samples_per_second": 1661.29, "characteristics.samples_per_second.normalized_per_core": 415.3225, "characteristics.samples_per_second.normalized_per_processor": 415.3225, "ck_system": "TitanRTXx4", "ck_used": false, "cooling": "watercooled", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) 8268", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/TitanRTXx4", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 107448, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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/TitanRTXx4", "system_name": "SCAN 3XS DBP T496X2 Fluid", "target_latency (ns)": 0, "target_qps": 1628, "task": "object detection", "task2": "object detection", "total_cores": 48, "uid": "4baaad281aa59daf", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32882294118, "90.00 percentile latency (ns)": 59132229510, "95.00 percentile latency (ns)": 62419244380, "97.00 percentile latency (ns)": 63702720143, "99.00 percentile latency (ns)": 65036336829, "99.90 percentile latency (ns)": 65595120120, "Max latency (ns)": 65793722371, "Mean latency (ns)": 32865761363, "Min duration satisfied": "Yes", "Min latency (ns)": 68568924, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2485.77, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "Shared with host", "accelerator_memory_configuration": "SRAM", "accelerator_model_name": "NVIDIA Xavier", "accelerator_on-chip_memories": "1MB (128KB/SM) L1 + 512KB L2 + 4MB (DLA)", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 22.926, "characteristics.samples_per_second": 2485.77, "characteristics.samples_per_second.normalized_per_core": 2485.77, "characteristics.samples_per_second.normalized_per_processor": 2485.77, "ck_system": "Xavier", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "JetPack 4.3 DP, TensorRT 6.0, cuDNN 7.6.3, CUDA 10.0, cub 1.8.0", "host_memory_capacity": "16 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "8MB L2 (2MB per dual cluster)/4 MB L3 (shared)", "host_processor_core_count": 8, "host_processor_frequency": "2265.5 MHz", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32 GB", "host_storage_type": "eMMC 5.1", "hw_notes": "GPU and both DLAs are used in Offline and MultiStream scenarios", "informal_model": "ssd-small", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/Xavier", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "pycuda 2019.1, pytorch 1.1, torchvision 0.2.2.post3", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 256, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 163548, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "frozen_inference_graph.pb", "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/Xavier", "system_name": "NVIDIA Jetson AGX Xavier", "target_latency (ns)": 0, "target_qps": 2478, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "5a10da6e80963554", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 248302143947, "90.00 percentile latency (ns)": 446843674201, "95.00 percentile latency (ns)": 471622790543, "97.00 percentile latency (ns)": 481559201342, "99.00 percentile latency (ns)": 491494910408, "99.90 percentile latency (ns)": 495920091502, "Max latency (ns)": 496486165893, "Mean latency (ns)": 248272510287, "Min duration satisfied": "Yes", "Min latency (ns)": 59450575, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 49.4999, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "Shared with host", "accelerator_memory_configuration": "SRAM", "accelerator_model_name": "NVIDIA Xavier", "accelerator_on-chip_memories": "1MB (128KB/SM) L1 + 512KB L2 + 4MB (DLA)", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 20.057, "characteristics.samples_per_second": 49.4999, "characteristics.samples_per_second.normalized_per_core": 49.4999, "characteristics.samples_per_second.normalized_per_processor": 49.4999, "ck_system": "Xavier", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "JetPack 4.3 DP, TensorRT 6.0, cuDNN 7.6.3, CUDA 10.0, cub 1.8.0", "host_memory_capacity": "16 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "8MB L2 (2MB per dual cluster)/4 MB L3 (shared)", "host_processor_core_count": 8, "host_processor_frequency": "2265.5 MHz", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32 GB", "host_storage_type": "eMMC 5.1", "hw_notes": "GPU and both DLAs are used in Offline and MultiStream scenarios", "informal_model": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/Xavier", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "pycuda 2019.1, pytorch 1.1, torchvision 0.2.2.post3", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 24576, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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/Xavier", "system_name": "NVIDIA Jetson AGX Xavier", "target_latency (ns)": 0, "target_qps": 50, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "f108617042580e4a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 34305726774, "90.00 percentile latency (ns)": 61917972007, "95.00 percentile latency (ns)": 65380104228, "97.00 percentile latency (ns)": 66771242096, "99.00 percentile latency (ns)": 68155573649, "99.90 percentile latency (ns)": 68779300476, "Max latency (ns)": 68848870687, "Mean latency (ns)": 34368119791, "Min duration satisfied": "Yes", "Min latency (ns)": 116920460, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 143084, "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 Tesla T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 143084, "characteristics.samples_per_second.normalized_per_core": 7154.2, "characteristics.samples_per_second.normalized_per_processor": 7154.2, "ck_system": "T4x20", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "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": "ssd-small", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/T4x20", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 256, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 9851160, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "frozen_inference_graph.pb", "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", "system_name": "Supermicro 6049GP-TRT-OTO-29 20xT4", "target_latency (ns)": 0, "target_qps": 149260, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "145a0a5376692458", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 31863080615, "90.00 percentile latency (ns)": 57541465186, "95.00 percentile latency (ns)": 60752974240, "97.00 percentile latency (ns)": 62040205616, "99.00 percentile latency (ns)": 63325326945, "99.90 percentile latency (ns)": 63900443315, "Max latency (ns)": 63955066872, "Mean latency (ns)": 31916684723, "Min duration satisfied": "Yes", "Min latency (ns)": 88756294, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2765.69, "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 Tesla T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 20.067, "characteristics.samples_per_second": 2765.69, "characteristics.samples_per_second.normalized_per_core": 138.2845, "characteristics.samples_per_second.normalized_per_processor": 138.2845, "ck_system": "T4x20", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "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": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/T4x20", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 176880, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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", "system_name": "Supermicro 6049GP-TRT-OTO-29 20xT4", "target_latency (ns)": 0, "target_qps": 2680, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "3e55f84852a43916", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32237545326, "90.00 percentile latency (ns)": 58217740073, "95.00 percentile latency (ns)": 61478437514, "97.00 percentile latency (ns)": 62786622311, "99.00 percentile latency (ns)": 64090969983, "99.90 percentile latency (ns)": 64671774306, "Max latency (ns)": 64734010021, "Mean latency (ns)": 32282204228, "Min duration satisfied": "Yes", "Min latency (ns)": 52891639, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 60871.6, "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 Tesla T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 60871.6, "characteristics.samples_per_second.normalized_per_core": 7608.95, "characteristics.samples_per_second.normalized_per_processor": 7608.95, "ck_system": "T4x8", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "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": "ssd-small", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/T4x8", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 256, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 3940464, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "frozen_inference_graph.pb", "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", "system_name": "Supermicro 4029GP-TRT-OTO-28 8xT4", "target_latency (ns)": 0, "target_qps": 59704, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "a699543f99551719", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32178954093, "90.00 percentile latency (ns)": 58108269585, "95.00 percentile latency (ns)": 61353496966, "97.00 percentile latency (ns)": 62659351625, "99.00 percentile latency (ns)": 63942731711, "99.90 percentile latency (ns)": 64541132014, "Max latency (ns)": 64607480650, "Mean latency (ns)": 32222073007, "Min duration satisfied": "Yes", "Min latency (ns)": 82755724, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1095.11, "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 Tesla T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.mAP": 20.067, "characteristics.samples_per_second": 1095.11, "characteristics.samples_per_second.normalized_per_core": 136.88875, "characteristics.samples_per_second.normalized_per_processor": 136.88875, "ck_system": "T4x8", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "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": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/T4x8", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 70752, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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", "system_name": "Supermicro 4029GP-TRT-OTO-28 8xT4", "target_latency (ns)": 0, "target_qps": 1072, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "a9de7c063f1eac63", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 31887151177, "90.00 percentile latency (ns)": 57470534496, "95.00 percentile latency (ns)": 60676992387, "97.00 percentile latency (ns)": 61946981994, "99.00 percentile latency (ns)": 63250849180, "99.90 percentile latency (ns)": 63815413060, "Max latency (ns)": 63863943442, "Mean latency (ns)": 31912274130, "Min duration satisfied": "Yes", "Min latency (ns)": 110229473, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 7750.85, "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.mAP": 20.111, "characteristics.samples_per_second": 7750.85, "characteristics.samples_per_second.normalized_per_core": 968.85625, "characteristics.samples_per_second.normalized_per_processor": 968.85625, "ck_system": "DGX-A100_A100-SXM4x8_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 495000, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7500, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "a9680b5f8b614d2d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 33076459328, "90.00 percentile latency (ns)": 59871573629, "95.00 percentile latency (ns)": 63247165577, "97.00 percentile latency (ns)": 64592696099, "99.00 percentile latency (ns)": 65938420773, "99.90 percentile latency (ns)": 66538682564, "Max latency (ns)": 66607160900, "Mean latency (ns)": 33160229866, "Min duration satisfied": "Yes", "Min latency (ns)": 80836625, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2774.48, "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.mAP": 20.111, "characteristics.samples_per_second": 2774.48, "characteristics.samples_per_second.normalized_per_core": 138.724, "characteristics.samples_per_second.normalized_per_processor": 138.724, "ck_system": "T4x20_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 184800, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 2800, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "7cd605423d8ec02d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 33719678815, "90.00 percentile latency (ns)": 60790200438, "95.00 percentile latency (ns)": 64164298835, "97.00 percentile latency (ns)": 65524285040, "99.00 percentile latency (ns)": 66886776667, "99.90 percentile latency (ns)": 67505566631, "Max latency (ns)": 67529908249, "Mean latency (ns)": 33735412373, "Min duration satisfied": "Yes", "Min latency (ns)": 158578613, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1688.85, "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.mAP": 20.111, "characteristics.samples_per_second": 1688.85, "characteristics.samples_per_second.normalized_per_core": 844.425, "characteristics.samples_per_second.normalized_per_processor": 844.425, "ck_system": "A100-PCIex2_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 114048, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 1728, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "75cd50c4e4ba50f9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 31719098742, "90.00 percentile latency (ns)": 57212626129, "95.00 percentile latency (ns)": 60389113008, "97.00 percentile latency (ns)": 61657466405, "99.00 percentile latency (ns)": 62943088452, "99.90 percentile latency (ns)": 63507434343, "Max latency (ns)": 63561625733, "Mean latency (ns)": 31750420557, "Min duration satisfied": "Yes", "Min latency (ns)": 67032994, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7787.72, "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.mAP": 20.111, "characteristics.samples_per_second": 7787.72, "characteristics.samples_per_second.normalized_per_core": 973.465, "characteristics.samples_per_second.normalized_per_processor": 973.465, "ck_system": "DGX-A100_A100-SXM4x8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 495000, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 7500, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "d28749a392e620a7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32934076506, "90.00 percentile latency (ns)": 59577103989, "95.00 percentile latency (ns)": 62916483692, "97.00 percentile latency (ns)": 64238140951, "99.00 percentile latency (ns)": 65578303476, "99.90 percentile latency (ns)": 66184920032, "Max latency (ns)": 66249571966, "Mean latency (ns)": 33012894105, "Min duration satisfied": "Yes", "Min latency (ns)": 80594832, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1111.8, "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.mAP": 20.111, "characteristics.samples_per_second": 1111.8, "characteristics.samples_per_second.normalized_per_core": 138.975, "characteristics.samples_per_second.normalized_per_processor": 138.975, "ck_system": "T4x8_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 73656, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 1116, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "41c9b30ee6571842", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 33350208247, "90.00 percentile latency (ns)": 60366225014, "95.00 percentile latency (ns)": 63745303622, "97.00 percentile latency (ns)": 65113242161, "99.00 percentile latency (ns)": 66469152364, "99.90 percentile latency (ns)": 67078599116, "Max latency (ns)": 67160358317, "Mean latency (ns)": 33431853846, "Min duration satisfied": "Yes", "Min latency (ns)": 88673682, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 2751.62, "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.mAP": 20.111, "characteristics.samples_per_second": 2751.62, "characteristics.samples_per_second.normalized_per_core": 137.581, "characteristics.samples_per_second.normalized_per_processor": 137.581, "ck_system": "T4x20_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 184800, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 2800, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "6bb7dff40b2aa23a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32985326381, "90.00 percentile latency (ns)": 59622533252, "95.00 percentile latency (ns)": 62957233557, "97.00 percentile latency (ns)": 64297502947, "99.00 percentile latency (ns)": 65636051794, "99.90 percentile latency (ns)": 66239157191, "Max latency (ns)": 66299890423, "Mean latency (ns)": 33043012682, "Min duration satisfied": "Yes", "Min latency (ns)": 80073590, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "TRTIS_Server", "Samples per second": 1110.95, "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.mAP": 20.111, "characteristics.samples_per_second": 1110.95, "characteristics.samples_per_second.normalized_per_core": 138.86875, "characteristics.samples_per_second.normalized_per_processor": 138.86875, "ck_system": "T4x8_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 73656, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 1116, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "5b97bf014a318ce9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 33876432112, "90.00 percentile latency (ns)": 61137093076, "95.00 percentile latency (ns)": 64559774309, "97.00 percentile latency (ns)": 65930405754, "99.00 percentile latency (ns)": 67321799666, "99.90 percentile latency (ns)": 67944230283, "Max latency (ns)": 67985032263, "Mean latency (ns)": 33918062281, "Min duration satisfied": "Yes", "Min latency (ns)": 81057353, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 1677.55, "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.mAP": 20.111, "characteristics.samples_per_second": 1677.55, "characteristics.samples_per_second.normalized_per_core": 838.775, "characteristics.samples_per_second.normalized_per_processor": 838.775, "ck_system": "A100-PCIex2_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "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": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 114048, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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": 1728, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "9469a89188011afa", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]