[ { "50.00 percentile": 1, "50.00 percentile latency (ns)": 30765018, "90.00 percentile": 1, "90.00 percentile latency (ns)": 47132496, "95.00 percentile": 1, "95.00 percentile latency (ns)": 47563921, "97.00 percentile": 1, "97.00 percentile latency (ns)": 47753214, "99.00 percentile": 1, "99.00 percentile latency (ns)": 48045198, "99.90 percentile": 2, "99.90 percentile latency (ns)": 48793366, "Max latency (ns)": 75834752, "Mean latency (ns)": 29039707, "Min duration satisfied": "Yes", "Min latency (ns)": 9574944, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per query": 1344, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.03, "characteristics.good": 38015, "characteristics.samples_per_query": 1344, "characteristics.samples_per_query.normalized_per_core": 1344.0, "characteristics.samples_per_query.normalized_per_processor": 1344.0, "characteristics.total": 50000, "ck_system": "A100-PCIex1_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/A100-PCIex1_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": 2048, "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": 1344, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "b9414c8a9bb4239e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 46596198, "90.00 percentile": 1, "90.00 percentile latency (ns)": 47895373, "95.00 percentile": 1, "95.00 percentile latency (ns)": 48095647, "97.00 percentile": 1, "97.00 percentile latency (ns)": 48214327, "99.00 percentile": 1, "99.00 percentile latency (ns)": 48424468, "99.90 percentile": 1, "99.90 percentile latency (ns)": 48770389, "Max latency (ns)": 61275061, "Mean latency (ns)": 46800106, "Min duration satisfied": "Yes", "Min latency (ns)": 44877689, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 96, "Scenario": "multistream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "Shared with host", "accelerator_memory_configuration": "SRAM", "accelerator_model_name": "NVIDIA AGX Xavier", "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.accuracy": 76.008, "characteristics.good": 38004, "characteristics.samples_per_query": 96, "characteristics.samples_per_query.normalized_per_core": 96.0, "characteristics.samples_per_query.normalized_per_processor": 96.0, "characteristics.total": 50000, "ck_system": "AGX_Xavier_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "21.03 Jetson CUDA-X AI Developer Preview, TensorRT 7.2.3, CUDA 10.2", "host_memory_capacity": "32 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 8, "host_processor_frequency": "", "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 resnet50, ssd-mobilenet, and ssd-resnet34, in Offline and MultiStream scenarios", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/AGX_Xavier_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "21.03 Jetson CUDA-X AI Developer Preview, TensorRT 7.2.3, CUDA 10.2, cuDNN 8.0.0, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 96, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 8, "uid": "4e288aed8db16efb", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 46228865, "90.00 percentile": 1, "90.00 percentile latency (ns)": 46511564, "95.00 percentile": 1, "95.00 percentile latency (ns)": 46594021, "97.00 percentile": 1, "97.00 percentile latency (ns)": 46655601, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47109151, "99.90 percentile": 2, "99.90 percentile latency (ns)": 52636725, "Max latency (ns)": 143563897, "Mean latency (ns)": 46196137, "Min duration satisfied": "Yes", "Min latency (ns)": 45115018, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 1760, "Scenario": "multistream", "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.accuracy": 76.03, "characteristics.good": 38015, "characteristics.samples_per_query": 1760, "characteristics.samples_per_query.normalized_per_core": 1760.0, "characteristics.samples_per_query.normalized_per_processor": 1760.0, "characteristics.total": 50000, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_edge", "ck_used": true, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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_edge", "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": 2048, "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": 1760, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx1_TRT_edge", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "d371d5e49ec98995", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 28528594, "90.00 percentile": 1, "90.00 percentile latency (ns)": 42163278, "95.00 percentile": 1, "95.00 percentile latency (ns)": 42425187, "97.00 percentile": 1, "97.00 percentile latency (ns)": 42560999, "99.00 percentile": 1, "99.00 percentile latency (ns)": 42770286, "99.90 percentile": 1, "99.90 percentile latency (ns)": 44047705, "Max latency (ns)": 408824619, "Mean latency (ns)": 25878031, "Min duration satisfied": "Yes", "Min latency (ns)": 8488453, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per query": 1344, "Scenario": "multistream", "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.accuracy": 76.03, "characteristics.good": 38015, "characteristics.samples_per_query": 1344, "characteristics.samples_per_query.normalized_per_core": 1344.0, "characteristics.samples_per_query.normalized_per_processor": 1344.0, "characteristics.total": 50000, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_Triton_edge", "ck_used": true, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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_edge", "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": 2048, "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": 1344, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx1_TRT_Triton_edge", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "b51c43e0930107d4", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 29956288, "90.00 percentile": 1, "90.00 percentile latency (ns)": 45330842, "95.00 percentile": 1, "95.00 percentile latency (ns)": 45709513, "97.00 percentile": 1, "97.00 percentile latency (ns)": 45950055, "99.00 percentile": 1, "99.00 percentile latency (ns)": 46220482, "99.90 percentile": 2, "99.90 percentile latency (ns)": 47986949, "Max latency (ns)": 71410309, "Mean latency (ns)": 30087659, "Min duration satisfied": "Yes", "Min latency (ns)": 8912922, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 1344, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.03, "characteristics.good": 38015, "characteristics.samples_per_query": 1344, "characteristics.samples_per_query.normalized_per_core": 1344.0, "characteristics.samples_per_query.normalized_per_processor": 1344.0, "characteristics.total": 50000, "ck_system": "A100-PCIex1_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/A100-PCIex1_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": 2048, "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": 1344, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex1_TRT", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "41888e4e2fdc0b4b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 46707363, "90.00 percentile": 1, "90.00 percentile latency (ns)": 47109630, "95.00 percentile": 1, "95.00 percentile latency (ns)": 47200643, "97.00 percentile": 1, "97.00 percentile latency (ns)": 47251257, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47373138, "99.90 percentile": 2, "99.90 percentile latency (ns)": 49218340, "Max latency (ns)": 77893277, "Mean latency (ns)": 44282031, "Min duration satisfied": "Yes", "Min latency (ns)": 29848643, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 50, "Scenario": "multistream", "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 NX", "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.accuracy": 76.048, "characteristics.good": 38024, "characteristics.samples_per_query": 50, "characteristics.samples_per_query.normalized_per_core": 50.0, "characteristics.samples_per_query.normalized_per_processor": 50.0, "characteristics.total": 50000, "ck_system": "Xavier_NX_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "21.03 Jetson CUDA-X AI Developer Preview, TensorRT 7.2.3, CUDA 10.2", "host_memory_capacity": "8 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 6, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32 GB", "host_storage_type": "Micro SD Card", "hw_notes": "GPU and both DLAs are used in resnet50, ssd-mobilenet, and ssd-resnet34, in Offline and MultiStream scenarios", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Xavier_NX_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "21.03 Jetson CUDA-X AI Developer Preview, TensorRT 7.2.3, CUDA 10.2, cuDNN 8.0.0, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 50, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Xavier_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 6, "uid": "823f0c1b8245f08b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 46725265, "90.00 percentile": 1, "90.00 percentile latency (ns)": 46977619, "95.00 percentile": 1, "95.00 percentile latency (ns)": 47011117, "97.00 percentile": 1, "97.00 percentile latency (ns)": 47067387, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47858538, "99.90 percentile": 2, "99.90 percentile latency (ns)": 67028216, "Max latency (ns)": 145839917, "Mean latency (ns)": 46785131, "Min duration satisfied": "Yes", "Min latency (ns)": 46400218, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 1760, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.92, "characteristics.good": 37960, "characteristics.samples_per_query": 1760, "characteristics.samples_per_query.normalized_per_core": 1760.0, "characteristics.samples_per_query.normalized_per_processor": 1760.0, "characteristics.total": 50000, "ck_system": "DGX-A100_A100-SXM4x1_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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-SXM4x1_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": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1760, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x1_TRT", "system_name": "NVIDIA DGX-A100 (1x A100-SXM4, TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "0bbc43ba670ebf67", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 30892977, "90.00 percentile": 1, "90.00 percentile latency (ns)": 46309189, "95.00 percentile": 1, "95.00 percentile latency (ns)": 46707613, "97.00 percentile": 1, "97.00 percentile latency (ns)": 46896510, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47194098, "99.90 percentile": 2, "99.90 percentile latency (ns)": 47751936, "Max latency (ns)": 75321115, "Mean latency (ns)": 28582696, "Min duration satisfied": "Yes", "Min latency (ns)": 9778327, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per query": 1344, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.92, "characteristics.good": 37960, "characteristics.samples_per_query": 1344, "characteristics.samples_per_query.normalized_per_core": 1344.0, "characteristics.samples_per_query.normalized_per_processor": 1344.0, "characteristics.total": 50000, "ck_system": "A100-PCIex1_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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-PCIex1_TRT_Triton", "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, Triton 20.09", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1344, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z52 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "065804ebdb44fff4", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 45897359, "90.00 percentile": 1, "90.00 percentile latency (ns)": 47137833, "95.00 percentile": 1, "95.00 percentile latency (ns)": 47296248, "97.00 percentile": 1, "97.00 percentile latency (ns)": 47389948, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47573603, "99.90 percentile": 1, "99.90 percentile latency (ns)": 47811357, "Max latency (ns)": 61289686, "Mean latency (ns)": 46116915, "Min duration satisfied": "Yes", "Min latency (ns)": 44472279, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 96, "Scenario": "multistream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "Shared with host", "accelerator_memory_configuration": "SRAM", "accelerator_model_name": "NVIDIA AGX Xavier", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 76.044, "characteristics.good": 38022, "characteristics.samples_per_query": 96, "characteristics.samples_per_query.normalized_per_core": 96.0, "characteristics.samples_per_query.normalized_per_processor": 96.0, "characteristics.total": 50000, "ck_system": "AGX_Xavier_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "20.09 Jetson CUDA-X AI Developer Preview, TensorRT 7.2, CUDA 10.2", "host_memory_capacity": "32GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 8, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32GB", "host_storage_type": "eMMC 5.1", "hw_notes": "GPU and both DLAs are used in resnet50, ssd-mobilenet, and ssd-resnet34, in Offline and MultiStream scenarios", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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/AGX_Xavier_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "20.09 Jetson CUDA-X AI Developer Preview, TensorRT 7.2, CUDA 10.2, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 96, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 8, "uid": "30854e21987af6a5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 32587659, "90.00 percentile": 1, "90.00 percentile latency (ns)": 46065628, "95.00 percentile": 1, "95.00 percentile latency (ns)": 46234376, "97.00 percentile": 1, "97.00 percentile latency (ns)": 46342419, "99.00 percentile": 1, "99.00 percentile latency (ns)": 46662055, "99.90 percentile": 2, "99.90 percentile latency (ns)": 51098381, "Max latency (ns)": 251558439, "Mean latency (ns)": 32665600, "Min duration satisfied": "Yes", "Min latency (ns)": 18235347, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per query": 1440, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.92, "characteristics.good": 37960, "characteristics.samples_per_query": 1440, "characteristics.samples_per_query.normalized_per_core": 1440.0, "characteristics.samples_per_query.normalized_per_processor": 1440.0, "characteristics.total": 50000, "ck_system": "DGX-A100_A100-SXM4x1_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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-SXM4x1_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": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1440, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x1_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "59c41fc8a8a6e27a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 36908603, "90.00 percentile": 1, "90.00 percentile latency (ns)": 46382358, "95.00 percentile": 1, "95.00 percentile latency (ns)": 46885569, "97.00 percentile": 1, "97.00 percentile latency (ns)": 47046630, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47250822, "99.90 percentile": 1, "99.90 percentile latency (ns)": 47637832, "Max latency (ns)": 81099797, "Mean latency (ns)": 34906764, "Min duration satisfied": "Yes", "Min latency (ns)": 19778042, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per query": 264, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.92, "characteristics.good": 37960, "characteristics.samples_per_query": 264, "characteristics.samples_per_query.normalized_per_core": 264.0, "characteristics.samples_per_query.normalized_per_processor": 264.0, "characteristics.total": 50000, "ck_system": "T4x1_TRT_Triton", "ck_used": true, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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/T4x1_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": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 264, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x1_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "204b4cb2c72844b7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 36554064, "90.00 percentile": 1, "90.00 percentile latency (ns)": 47194125, "95.00 percentile": 1, "95.00 percentile latency (ns)": 47720085, "97.00 percentile": 1, "97.00 percentile latency (ns)": 47824155, "99.00 percentile": 1, "99.00 percentile latency (ns)": 47975303, "99.90 percentile": 1, "99.90 percentile latency (ns)": 48297153, "Max latency (ns)": 78726461, "Mean latency (ns)": 35267245, "Min duration satisfied": "Yes", "Min latency (ns)": 19414934, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 272, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.92, "characteristics.good": 37960, "characteristics.samples_per_query": 272, "characteristics.samples_per_query.normalized_per_core": 272.0, "characteristics.samples_per_query.normalized_per_processor": 272.0, "characteristics.total": 50000, "ck_system": "T4x1_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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/T4x1_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": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 272, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x T4, TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 56, "uid": "e6e229d34c494f82", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 29016969, "90.00 percentile": 1, "90.00 percentile latency (ns)": 43971526, "95.00 percentile": 1, "95.00 percentile latency (ns)": 44308315, "97.00 percentile": 1, "97.00 percentile latency (ns)": 44488286, "99.00 percentile": 1, "99.00 percentile latency (ns)": 44776775, "99.90 percentile": 1, "99.90 percentile latency (ns)": 45795757, "Max latency (ns)": 69270060, "Mean latency (ns)": 29171495, "Min duration satisfied": "Yes", "Min latency (ns)": 13213220, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 1344, "Scenario": "multistream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.92, "characteristics.good": 37960, "characteristics.samples_per_query": 1344, "characteristics.samples_per_query.normalized_per_core": 1344.0, "characteristics.samples_per_query.normalized_per_processor": 1344.0, "characteristics.total": 50000, "ck_system": "A100-PCIex1_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "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": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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-PCIex1_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": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1344, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex1_TRT", "system_name": "Gigabyte G482-Z52 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "4385999bfdb49f5c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile": 1, "50.00 percentile latency (ns)": 48490441, "90.00 percentile": 1, "90.00 percentile latency (ns)": 49419151, "95.00 percentile": 1, "95.00 percentile latency (ns)": 49541788, "97.00 percentile": 1, "97.00 percentile latency (ns)": 49600254, "99.00 percentile": 1, "99.00 percentile latency (ns)": 49749181, "99.90 percentile": 2, "99.90 percentile latency (ns)": 51472210, "Max latency (ns)": 86152203, "Mean latency (ns)": 45280331, "Min duration satisfied": "Yes", "Min latency (ns)": 35312561, "Min queries satisfied": "Yes", "Mode": "Performance", "Per-sample latency": "", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per query": 50, "Scenario": "multistream", "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 NX", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 76.01, "characteristics.good": 38005, "characteristics.samples_per_query": 50, "characteristics.samples_per_query.normalized_per_core": 50.0, "characteristics.samples_per_query.normalized_per_processor": 50.0, "characteristics.total": 50000, "ck_system": "Xavier_NX_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_query", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "20.09 Jetson CUDA-X AI Developer Preview, TensorRT 7.2, CUDA 10.2", "host_memory_capacity": "8GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 6, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32GB", "host_storage_type": "Micro SD Card", "hw_notes": "GPU and both DLAs are used in resnet50, ssd-mobilenet, and ssd-resnet34, in Offline and MultiStream scenarios", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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/Xavier_NX_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "20.09 Jetson CUDA-X AI Developer Preview, TensorRT 7.2, CUDA 10.2, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2048, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 50, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Xavier_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 50000000, "target_qps": 20, "task": "image classification", "task2": "image classification", "total_cores": 6, "uid": "041418930094e815", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]