[ { "50.00 percentile latency (ns)": 321216051841, "90.00 percentile latency (ns)": 578253679943, "95.00 percentile latency (ns)": 610410962110, "97.00 percentile latency (ns)": 623251451278, "99.00 percentile latency (ns)": 636131414805, "99.90 percentile latency (ns)": 641908543009, "Max latency (ns)": 642520234022, "Mean latency (ns)": 321206938116, "Min duration satisfied": "Yes", "Min latency (ns)": 76625697, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 24652.9, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 24652.9, "characteristics.samples_per_second.normalized_per_core": 24652.9, "characteristics.samples_per_second.normalized_per_processor": 24652.9, "ck_system": "A30x1_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-mobilenet", "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/A30x1_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": 1024, "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": 15840000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A30x1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 24000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "c00974bd4c1f9e2a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325282821729, "90.00 percentile latency (ns)": 585774527028, "95.00 percentile latency (ns)": 618356031691, "97.00 percentile latency (ns)": 631321470318, "99.00 percentile latency (ns)": 644356328005, "99.90 percentile latency (ns)": 650267920554, "Max latency (ns)": 650856273030, "Mean latency (ns)": 325305745346, "Min duration satisfied": "Yes", "Min latency (ns)": 66399074, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 476.603, "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": 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": 476.603, "characteristics.samples_per_second.normalized_per_core": 476.603, "characteristics.samples_per_second.normalized_per_processor": 476.603, "ck_system": "A30x1_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": 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/A30x1_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": 310200, "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/A30x1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 470, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "68b6b474a9895ee0", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 337893776946, "90.00 percentile latency (ns)": 607946579728, "95.00 percentile latency (ns)": 641697185471, "97.00 percentile latency (ns)": 655154280727, "99.00 percentile latency (ns)": 668686310434, "99.90 percentile latency (ns)": 674739148923, "Max latency (ns)": 675427975381, "Mean latency (ns)": 337898569548, "Min duration satisfied": "Yes", "Min latency (ns)": 259722210, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1495.05, "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 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.mAP": 22.88, "characteristics.power": 19.798700740740724, "characteristics.power.normalized_per_core": 19.798700740740724, "characteristics.power.normalized_per_processor": 19.798700740740724, "characteristics.samples_per_second": 1495.05, "characteristics.samples_per_second.normalized_per_core": 1495.05, "characteristics.samples_per_second.normalized_per_processor": 1495.05, "ck_system": "Xavier_NX_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "informal_model": "ssd-mobilenet", "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/Xavier_NX_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "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": 1009800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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_NX_TRT_MaxQ", "system_name": "Auvidea JNX30 Xavier NX (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1530, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "be502d27ef634712", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 441560918926, "90.00 percentile latency (ns)": 794933560710, "95.00 percentile latency (ns)": 839130360187, "97.00 percentile latency (ns)": 856713174374, "99.00 percentile latency (ns)": 874382255355, "99.90 percentile latency (ns)": 882292082300, "Max latency (ns)": 883130426746, "Mean latency (ns)": 441524704170, "Min duration satisfied": "Yes", "Min latency (ns)": 60954922, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 29.8937, "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 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.mAP": 19.973, "characteristics.power": 19.777793884484712, "characteristics.power.normalized_per_core": 19.777793884484712, "characteristics.power.normalized_per_processor": 19.777793884484712, "characteristics.samples_per_second": 29.8937, "characteristics.samples_per_second.normalized_per_core": 29.8937, "characteristics.samples_per_second.normalized_per_processor": 29.8937, "ck_system": "Xavier_NX_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "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/Xavier_NX_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, 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": 26400, "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/Xavier_NX_TRT_MaxQ", "system_name": "Auvidea JNX30 Xavier NX (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 40, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "6524909b097f731a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 332570661385, "90.00 percentile latency (ns)": 598357713972, "95.00 percentile latency (ns)": 631622743589, "97.00 percentile latency (ns)": 644861138036, "99.00 percentile latency (ns)": 658167085138, "99.90 percentile latency (ns)": 664156604506, "Max latency (ns)": 664772163523, "Mean latency (ns)": 332469632085, "Min duration satisfied": "Yes", "Min latency (ns)": 65261095, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 2432.41, "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 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.mAP": 22.914, "characteristics.samples_per_second": 2432.41, "characteristics.samples_per_second.normalized_per_core": 2432.41, "characteristics.samples_per_second.normalized_per_processor": 2432.41, "ck_system": "AGX_Xavier_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "informal_model": "ssd-mobilenet", "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/AGX_Xavier_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "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": 1617000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/AGX_Xavier_TRT_Triton", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 2450, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "6d6e906877fe4a67", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 337122216087, "90.00 percentile latency (ns)": 630938588251, "95.00 percentile latency (ns)": 677837426241, "97.00 percentile latency (ns)": 696646361792, "99.00 percentile latency (ns)": 715482067680, "99.90 percentile latency (ns)": 723952063549, "Max latency (ns)": 724875900332, "Mean latency (ns)": 341946356290, "Min duration satisfied": "Yes", "Min latency (ns)": 171807515, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "DLA_Triton_Server", "Samples per second": 50.988, "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 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.mAP": 19.981, "characteristics.samples_per_second": 50.988, "characteristics.samples_per_second.normalized_per_core": 50.988, "characteristics.samples_per_second.normalized_per_processor": 50.988, "ck_system": "AGX_Xavier_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "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/AGX_Xavier_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, 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": 36960, "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/AGX_Xavier_TRT_Triton", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 56, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "92e5a47c1b42a8b9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 339937284881, "90.00 percentile latency (ns)": 611458994098, "95.00 percentile latency (ns)": 645385768731, "97.00 percentile latency (ns)": 658950966289, "99.00 percentile latency (ns)": 672516837454, "99.90 percentile latency (ns)": 678642807526, "Max latency (ns)": 679298822946, "Mean latency (ns)": 339730904217, "Min duration satisfied": "Yes", "Min latency (ns)": 57409755, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 42750, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 42750, "characteristics.samples_per_second.normalized_per_core": 42750.0, "characteristics.samples_per_second.normalized_per_processor": 42750.0, "ck_system": "A100-PCIex1_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-mobilenet", "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/A100-PCIex1_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": 1024, "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": 29040000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 44000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "cee1a309ce4ecbae", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 385404399123, "90.00 percentile latency (ns)": 694230049255, "95.00 percentile latency (ns)": 732823266382, "97.00 percentile latency (ns)": 748267960940, "99.00 percentile latency (ns)": 763708208393, "99.90 percentile latency (ns)": 770648529409, "Max latency (ns)": 771349698545, "Mean latency (ns)": 385405603439, "Min duration satisfied": "Yes", "Min latency (ns)": 81366650, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 821.417, "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": 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": 821.417, "characteristics.samples_per_second.normalized_per_core": 821.417, "characteristics.samples_per_second.normalized_per_processor": 821.417, "ck_system": "A100-PCIex1_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": 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/A100-PCIex1_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": 633600, "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-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "93cb6a18b0826e84", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328592501698, "90.00 percentile latency (ns)": 591269859420, "95.00 percentile latency (ns)": 624069948896, "97.00 percentile latency (ns)": 637165416091, "99.00 percentile latency (ns)": 650322486528, "99.90 percentile latency (ns)": 656226763406, "Max latency (ns)": 656834347239, "Mean latency (ns)": 328549238541, "Min duration satisfied": "Yes", "Min latency (ns)": 88404708, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2833.59, "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 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.mAP": 22.899, "characteristics.samples_per_second": 2833.59, "characteristics.samples_per_second.normalized_per_core": 2833.59, "characteristics.samples_per_second.normalized_per_processor": 2833.59, "ck_system": "AGX_Xavier_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "informal_model": "ssd-mobilenet", "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/AGX_Xavier_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "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": 1861200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 2820, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "f92d5a4922910bc8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 335049669373, "90.00 percentile latency (ns)": 603080645854, "95.00 percentile latency (ns)": 636497061144, "97.00 percentile latency (ns)": 649947384807, "99.00 percentile latency (ns)": 663366362073, "99.90 percentile latency (ns)": 669460729990, "Max latency (ns)": 670104496963, "Mean latency (ns)": 335015629289, "Min duration satisfied": "Yes", "Min latency (ns)": 52666722, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 55.1556, "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 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.mAP": 19.967, "characteristics.samples_per_second": 55.1556, "characteristics.samples_per_second.normalized_per_core": 55.1556, "characteristics.samples_per_second.normalized_per_processor": 55.1556, "ck_system": "AGX_Xavier_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "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/AGX_Xavier_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, 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": 36960, "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/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 56, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "b7e444d7a1d8c444", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 336130150347, "90.00 percentile latency (ns)": 605011263853, "95.00 percentile latency (ns)": 638605237948, "97.00 percentile latency (ns)": 652050280304, "99.00 percentile latency (ns)": 665495128258, "99.90 percentile latency (ns)": 671528917346, "Max latency (ns)": 672194826602, "Mean latency (ns)": 336126854175, "Min duration satisfied": "Yes", "Min latency (ns)": 38824154, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 6873.01, "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": 22.914, "characteristics.samples_per_second": 6873.01, "characteristics.samples_per_second.normalized_per_core": 6873.01, "characteristics.samples_per_second.normalized_per_processor": 6873.01, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.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-mobilenet", "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_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": 1024, "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": 4620000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_Triton", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 7000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "4cce94844f7287dd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330877321755, "90.00 percentile latency (ns)": 595208223405, "95.00 percentile latency (ns)": 628455979484, "97.00 percentile latency (ns)": 641518634693, "99.00 percentile latency (ns)": 654819448952, "99.90 percentile latency (ns)": 660757286412, "Max latency (ns)": 661321346366, "Mean latency (ns)": 330799366877, "Min duration satisfied": "Yes", "Min latency (ns)": 239801855, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 134.73, "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": 134.73, "characteristics.samples_per_second.normalized_per_core": 134.73, "characteristics.samples_per_second.normalized_per_processor": 134.73, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.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": 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_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": 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_Triton", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 135, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "fddd0639480da582", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 303541853618, "90.00 percentile latency (ns)": 546970483399, "95.00 percentile latency (ns)": 577367553874, "97.00 percentile latency (ns)": 589550809490, "99.00 percentile latency (ns)": 601710411240, "99.90 percentile latency (ns)": 607209587321, "Max latency (ns)": 607795523634, "Mean latency (ns)": 303623587636, "Min duration satisfied": "Yes", "Min latency (ns)": 237075445, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 47779.2, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 47779.2, "characteristics.samples_per_second.normalized_per_core": 47779.2, "characteristics.samples_per_second.normalized_per_processor": 47779.2, "ck_system": "A100-PCIe-80GBx1_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-mobilenet", "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/A100-PCIe-80GBx1_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": 1024, "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": 29040000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIe-80GBx1_TRT", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe-80GB, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 44000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "019c92a7d4b4b955", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 346558201452, "90.00 percentile latency (ns)": 624482176852, "95.00 percentile latency (ns)": 659151788554, "97.00 percentile latency (ns)": 673053213186, "99.00 percentile latency (ns)": 686956439265, "99.90 percentile latency (ns)": 693252707406, "Max latency (ns)": 693817756287, "Mean latency (ns)": 346577154648, "Min duration satisfied": "Yes", "Min latency (ns)": 145895066, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 913.208, "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": 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": 913.208, "characteristics.samples_per_second.normalized_per_core": 913.208, "characteristics.samples_per_second.normalized_per_processor": 913.208, "ck_system": "A100-PCIe-80GBx1_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": 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/A100-PCIe-80GBx1_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": 633600, "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-80GBx1_TRT", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe-80GB, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "4d2c814c88a80654", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326242128362, "90.00 percentile latency (ns)": 587362849211, "95.00 percentile latency (ns)": 620017856828, "97.00 percentile latency (ns)": 633077042633, "99.00 percentile latency (ns)": 646134441065, "99.90 percentile latency (ns)": 652008395604, "Max latency (ns)": 652645291001, "Mean latency (ns)": 326269566279, "Min duration satisfied": "Yes", "Min latency (ns)": 228259536, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 51777, "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": 22.914, "characteristics.samples_per_second": 51777, "characteristics.samples_per_second.normalized_per_core": 51777.0, "characteristics.samples_per_second.normalized_per_processor": 51777.0, "ck_system": "DGX-A100_A100-SXM-80GBx1_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-mobilenet", "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-80GBx1_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": 1024, "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": 33792000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM-80GBx1_TRT", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "946d798cdd4a3eec", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 318616351408, "90.00 percentile latency (ns)": 573608990186, "95.00 percentile latency (ns)": 605414775135, "97.00 percentile latency (ns)": 618166875793, "99.00 percentile latency (ns)": 630913062590, "99.90 percentile latency (ns)": 636711402829, "Max latency (ns)": 637226849099, "Mean latency (ns)": 318571399586, "Min duration satisfied": "Yes", "Min latency (ns)": 129309838, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 994.308, "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": 20.117, "characteristics.samples_per_second": 994.308, "characteristics.samples_per_second.normalized_per_core": 994.308, "characteristics.samples_per_second.normalized_per_processor": 994.308, "ck_system": "DGX-A100_A100-SXM-80GBx1_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": 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-80GBx1_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": 633600, "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-80GBx1_TRT", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "e17f24a6284cb6b7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330981390748, "90.00 percentile latency (ns)": 594796152532, "95.00 percentile latency (ns)": 627595228278, "97.00 percentile latency (ns)": 640925987894, "99.00 percentile latency (ns)": 654297534020, "99.90 percentile latency (ns)": 660231143934, "Max latency (ns)": 660857249710, "Mean latency (ns)": 330816845492, "Min duration satisfied": "Yes", "Min latency (ns)": 137377030, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 1498.05, "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 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.mAP": 22.914, "characteristics.samples_per_second": 1498.05, "characteristics.samples_per_second.normalized_per_core": 1498.05, "characteristics.samples_per_second.normalized_per_processor": 1498.05, "ck_system": "Xavier_NX_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "informal_model": "ssd-mobilenet", "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/Xavier_NX_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "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": 990000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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_NX_TRT_Triton", "system_name": "NVIDIA Jetson Xavier NX (TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1500, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "bc13fd43849567d9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 353767394799, "90.00 percentile latency (ns)": 636306258822, "95.00 percentile latency (ns)": 671624939325, "97.00 percentile latency (ns)": 693769157259, "99.00 percentile latency (ns)": 715914564978, "99.90 percentile latency (ns)": 725852136330, "Max latency (ns)": 726942239995, "Mean latency (ns)": 354243927785, "Min duration satisfied": "Yes", "Min latency (ns)": 415292860, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "DLA_Triton_Server", "Samples per second": 36.3165, "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 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.mAP": 19.917, "characteristics.samples_per_second": 36.3165, "characteristics.samples_per_second.normalized_per_core": 36.3165, "characteristics.samples_per_second.normalized_per_processor": 36.3165, "ck_system": "Xavier_NX_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "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/Xavier_NX_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, 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": 26400, "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/Xavier_NX_TRT_Triton", "system_name": "NVIDIA Jetson Xavier NX (TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 40, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "284b2595b30936ff", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330435265810, "90.00 percentile latency (ns)": 594863193421, "95.00 percentile latency (ns)": 627912616252, "97.00 percentile latency (ns)": 641135735043, "99.00 percentile latency (ns)": 654368829245, "99.90 percentile latency (ns)": 660315708534, "Max latency (ns)": 660975266574, "Mean latency (ns)": 330431053435, "Min duration satisfied": "Yes", "Min latency (ns)": 45795595, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6290.7, "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": 22.914, "characteristics.samples_per_second": 6290.7, "characteristics.samples_per_second.normalized_per_core": 6290.7, "characteristics.samples_per_second.normalized_per_processor": 6290.7, "ck_system": "A30-MIG_1x1g.6gb_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-mobilenet", "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", "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": 1024, "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": 4158000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A30-MIG_1x1g.6gb_TRT", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 6300, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "f40d28fbe9ef1cf8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331142386350, "90.00 percentile latency (ns)": 596080479931, "95.00 percentile latency (ns)": 629203586239, "97.00 percentile latency (ns)": 642445320020, "99.00 percentile latency (ns)": 655680468481, "99.90 percentile latency (ns)": 661640867625, "Max latency (ns)": 662300143532, "Mean latency (ns)": 331126918438, "Min duration satisfied": "Yes", "Min latency (ns)": 32375698, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 127.555, "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": 127.555, "characteristics.samples_per_second.normalized_per_core": 127.555, "characteristics.samples_per_second.normalized_per_processor": 127.555, "ck_system": "A30-MIG_1x1g.6gb_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": 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", "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": 84480, "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", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 128, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "43f2bd1b3cb172f2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 349548602728, "90.00 percentile latency (ns)": 629098339029, "95.00 percentile latency (ns)": 664080993854, "97.00 percentile latency (ns)": 678084896227, "99.00 percentile latency (ns)": 692034006563, "99.90 percentile latency (ns)": 698313404428, "Max latency (ns)": 698972905458, "Mean latency (ns)": 349595033783, "Min duration satisfied": "Yes", "Min latency (ns)": 201569692, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1869.6, "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 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.mAP": 22.873, "characteristics.power": 23.935343839541552, "characteristics.power.normalized_per_core": 23.935343839541552, "characteristics.power.normalized_per_processor": 23.935343839541552, "characteristics.samples_per_second": 1869.6, "characteristics.samples_per_second.normalized_per_core": 1869.6, "characteristics.samples_per_second.normalized_per_processor": 1869.6, "ck_system": "AGX_Xavier_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "informal_model": "ssd-mobilenet", "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/AGX_Xavier_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "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": 1306800, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/AGX_Xavier_TRT_MaxQ", "system_name": "Auvidea X220-LC AGX Xavier 32GB (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1980, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "b84ec48b59037967", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 354238635535, "90.00 percentile latency (ns)": 637659274510, "95.00 percentile latency (ns)": 673094965252, "97.00 percentile latency (ns)": 687268573488, "99.00 percentile latency (ns)": 701371964862, "99.90 percentile latency (ns)": 707804144066, "Max latency (ns)": 708490595307, "Mean latency (ns)": 354163661174, "Min duration satisfied": "Yes", "Min latency (ns)": 77128291, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 39.1254, "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 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.mAP": 19.978, "characteristics.power": 24.257212976022565, "characteristics.power.normalized_per_core": 24.257212976022565, "characteristics.power.normalized_per_processor": 24.257212976022565, "characteristics.samples_per_second": 39.1254, "characteristics.samples_per_second.normalized_per_core": 39.1254, "characteristics.samples_per_second.normalized_per_processor": 39.1254, "ck_system": "AGX_Xavier_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "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/AGX_Xavier_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, 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": 27720, "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/AGX_Xavier_TRT_MaxQ", "system_name": "Auvidea X220-LC AGX Xavier 32GB (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 42, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "99bd744efb0b8d07", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 304689596732, "90.00 percentile latency (ns)": 548707733676, "95.00 percentile latency (ns)": 579205823585, "97.00 percentile latency (ns)": 591405102178, "99.00 percentile latency (ns)": 603594790361, "99.90 percentile latency (ns)": 609101147654, "Max latency (ns)": 609690338964, "Mean latency (ns)": 304690169155, "Min duration satisfied": "Yes", "Min latency (ns)": 36346918, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 47630.7, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 47630.7, "characteristics.samples_per_second.normalized_per_core": 47630.7, "characteristics.samples_per_second.normalized_per_processor": 47630.7, "ck_system": "A100-PCIe-80GBx1_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-mobilenet", "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/A100-PCIe-80GBx1_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": 1024, "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": 29040000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIe-80GBx1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe-80GB, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 44000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "a9b4aac16384c492", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 347978179690, "90.00 percentile latency (ns)": 626799937416, "95.00 percentile latency (ns)": 661640876889, "97.00 percentile latency (ns)": 675577673892, "99.00 percentile latency (ns)": 689515134317, "99.90 percentile latency (ns)": 695780110875, "Max latency (ns)": 696413206949, "Mean latency (ns)": 347974079695, "Min duration satisfied": "Yes", "Min latency (ns)": 76498926, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 909.805, "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": 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": 909.805, "characteristics.samples_per_second.normalized_per_core": 909.805, "characteristics.samples_per_second.normalized_per_processor": 909.805, "ck_system": "A100-PCIe-80GBx1_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": 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/A100-PCIe-80GBx1_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": 633600, "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-80GBx1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe-80GB, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "3c474799c7408348", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 338251189246, "90.00 percentile latency (ns)": 609103222524, "95.00 percentile latency (ns)": 642939139586, "97.00 percentile latency (ns)": 656501045216, "99.00 percentile latency (ns)": 670021399802, "99.90 percentile latency (ns)": 676116419950, "Max latency (ns)": 676785806087, "Mean latency (ns)": 338216436840, "Min duration satisfied": "Yes", "Min latency (ns)": 60513784, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 18528.8, "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 A10", "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": 22.914, "characteristics.samples_per_second": 18528.8, "characteristics.samples_per_second.normalized_per_core": 18528.8, "characteristics.samples_per_second.normalized_per_processor": 18528.8, "ck_system": "A10x1_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-mobilenet", "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/A10x1_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": 1024, "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": 12540000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A10x1_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 19000, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "eb9bf7db4da5b7df", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 322626709281, "90.00 percentile latency (ns)": 582148767312, "95.00 percentile latency (ns)": 614623280954, "97.00 percentile latency (ns)": 627518237234, "99.00 percentile latency (ns)": 640521619045, "99.90 percentile latency (ns)": 646413091216, "Max latency (ns)": 647001376719, "Mean latency (ns)": 322840366491, "Min duration satisfied": "Yes", "Min latency (ns)": 101386734, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 316.228, "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": 1, "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": 316.228, "characteristics.samples_per_second.normalized_per_core": 316.228, "characteristics.samples_per_second.normalized_per_processor": 316.228, "ck_system": "A10x1_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": 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/A10x1_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": 204600, "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/A10x1_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 310, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "41a2b9cbe7215065", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328127919212, "90.00 percentile latency (ns)": 590800922264, "95.00 percentile latency (ns)": 623633396783, "97.00 percentile latency (ns)": 636766094695, "99.00 percentile latency (ns)": 649899688527, "99.90 percentile latency (ns)": 655810534954, "Max latency (ns)": 656449793583, "Mean latency (ns)": 328143564892, "Min duration satisfied": "Yes", "Min latency (ns)": 41311325, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 51476.9, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 51476.9, "characteristics.samples_per_second.normalized_per_core": 51476.9, "characteristics.samples_per_second.normalized_per_processor": 51476.9, "ck_system": "DGX-A100_A100-SXM-80GBx1_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-mobilenet", "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-80GBx1_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": 1024, "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": 33792000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM-80GBx1_TRT_Triton", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "6520ad84084649ae", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 319928569039, "90.00 percentile latency (ns)": 576130361196, "95.00 percentile latency (ns)": 608156754214, "97.00 percentile latency (ns)": 620964964637, "99.00 percentile latency (ns)": 633777200037, "99.90 percentile latency (ns)": 639535006257, "Max latency (ns)": 640118066495, "Mean latency (ns)": 319924320839, "Min duration satisfied": "Yes", "Min latency (ns)": 64004329, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 989.817, "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": 20.117, "characteristics.samples_per_second": 989.817, "characteristics.samples_per_second.normalized_per_core": 989.817, "characteristics.samples_per_second.normalized_per_processor": 989.817, "ck_system": "DGX-A100_A100-SXM-80GBx1_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": 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-80GBx1_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": 633600, "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-80GBx1_TRT_Triton", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "94ca7fcc7ad63962", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 338090277935, "90.00 percentile latency (ns)": 609055242490, "95.00 percentile latency (ns)": 642911582739, "97.00 percentile latency (ns)": 656478906874, "99.00 percentile latency (ns)": 670007194525, "99.90 percentile latency (ns)": 676103490081, "Max latency (ns)": 676770301645, "Mean latency (ns)": 338118091065, "Min duration satisfied": "Yes", "Min latency (ns)": 194699319, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 18529.2, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 18529.2, "characteristics.samples_per_second.normalized_per_core": 18529.2, "characteristics.samples_per_second.normalized_per_processor": 18529.2, "ck_system": "A10x1_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-mobilenet", "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/A10x1_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": 1024, "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": 12540000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A10x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 19000, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "c44d97a46d964611", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 322445146925, "90.00 percentile latency (ns)": 581714142706, "95.00 percentile latency (ns)": 614148866649, "97.00 percentile latency (ns)": 627030187201, "99.00 percentile latency (ns)": 640010274410, "99.90 percentile latency (ns)": 645890419274, "Max latency (ns)": 646471320108, "Mean latency (ns)": 322624726051, "Min duration satisfied": "Yes", "Min latency (ns)": 102632963, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 316.487, "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": 1, "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": 316.487, "characteristics.samples_per_second.normalized_per_core": 316.487, "characteristics.samples_per_second.normalized_per_processor": 316.487, "ck_system": "A10x1_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": 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/A10x1_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": 204600, "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/A10x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 310, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "62af8478df2f4dc0", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325299366862, "90.00 percentile latency (ns)": 585513655295, "95.00 percentile latency (ns)": 618092294174, "97.00 percentile latency (ns)": 631057421864, "99.00 percentile latency (ns)": 644090794358, "99.90 percentile latency (ns)": 649971234376, "Max latency (ns)": 650549716941, "Mean latency (ns)": 325309267980, "Min duration satisfied": "Yes", "Min latency (ns)": 227144096, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 24348.6, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 24348.6, "characteristics.samples_per_second.normalized_per_core": 24348.6, "characteristics.samples_per_second.normalized_per_processor": 24348.6, "ck_system": "A30x1_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-mobilenet", "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/A30x1_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": 1024, "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": 15840000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A30x1_TRT", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 24000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "4736b84e82b70e4b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325067217452, "90.00 percentile latency (ns)": 585188088926, "95.00 percentile latency (ns)": 617660338196, "97.00 percentile latency (ns)": 630671292917, "99.00 percentile latency (ns)": 643679294998, "99.90 percentile latency (ns)": 649570017915, "Max latency (ns)": 650089411355, "Mean latency (ns)": 325057325638, "Min duration satisfied": "Yes", "Min latency (ns)": 138110931, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 477.165, "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": 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": 477.165, "characteristics.samples_per_second.normalized_per_core": 477.165, "characteristics.samples_per_second.normalized_per_processor": 477.165, "ck_system": "A30x1_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": 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/A30x1_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": 310200, "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/A30x1_TRT", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 470, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "7fa664b5c2a4416b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 324303919517, "90.00 percentile latency (ns)": 583695309780, "95.00 percentile latency (ns)": 616105971748, "97.00 percentile latency (ns)": 629077289014, "99.00 percentile latency (ns)": 642048511610, "99.90 percentile latency (ns)": 647869628703, "Max latency (ns)": 648512148120, "Mean latency (ns)": 324296541629, "Min duration satisfied": "Yes", "Min latency (ns)": 62944135, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7124, "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": 22.914, "characteristics.samples_per_second": 7124, "characteristics.samples_per_second.normalized_per_core": 7124.0, "characteristics.samples_per_second.normalized_per_processor": 7124.0, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_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-mobilenet", "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", "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": 1024, "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": 4620000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 7000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "4d9ab4b1dddaca15", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331642440860, "90.00 percentile latency (ns)": 596167737194, "95.00 percentile latency (ns)": 629475830732, "97.00 percentile latency (ns)": 642796936904, "99.00 percentile latency (ns)": 656118935745, "99.90 percentile latency (ns)": 661830309653, "Max latency (ns)": 662396234794, "Mean latency (ns)": 331451013297, "Min duration satisfied": "Yes", "Min latency (ns)": 476483989, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 134.512, "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": 134.512, "characteristics.samples_per_second.normalized_per_core": 134.512, "characteristics.samples_per_second.normalized_per_processor": 134.512, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_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": 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", "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", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 135, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "0801490b65e24a14", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327675775877, "90.00 percentile latency (ns)": 589578193757, "95.00 percentile latency (ns)": 622231626845, "97.00 percentile latency (ns)": 635294717946, "99.00 percentile latency (ns)": 648361555797, "99.90 percentile latency (ns)": 654255859539, "Max latency (ns)": 654901836819, "Mean latency (ns)": 327613726073, "Min duration satisfied": "Yes", "Min latency (ns)": 21701206, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 6852.94, "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": 22.914, "characteristics.samples_per_second": 6852.94, "characteristics.samples_per_second.normalized_per_core": 6852.94, "characteristics.samples_per_second.normalized_per_processor": 6852.94, "ck_system": "A30-MIG_1x1g.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-mobilenet", "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_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": 1024, "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": 4488000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A30-MIG_1x1g.6gb_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 6800, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "d914aeec4284445a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 323822216298, "90.00 percentile latency (ns)": 583191250096, "95.00 percentile latency (ns)": 615573899766, "97.00 percentile latency (ns)": 628542387829, "99.00 percentile latency (ns)": 641510847810, "99.90 percentile latency (ns)": 647344652122, "Max latency (ns)": 647979793216, "Mean latency (ns)": 323880489683, "Min duration satisfied": "Yes", "Min latency (ns)": 31494384, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 132.412, "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": 132.412, "characteristics.samples_per_second.normalized_per_core": 132.412, "characteristics.samples_per_second.normalized_per_processor": 132.412, "ck_system": "A30-MIG_1x1g.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": 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_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": 85800, "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_Triton", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 130, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "1fdf48ba81e63ac3", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 339923149768, "90.00 percentile latency (ns)": 611984142966, "95.00 percentile latency (ns)": 646005292903, "97.00 percentile latency (ns)": 659628170922, "99.00 percentile latency (ns)": 673209461975, "99.90 percentile latency (ns)": 679351781451, "Max latency (ns)": 679995995688, "Mean latency (ns)": 339976858693, "Min duration satisfied": "Yes", "Min latency (ns)": 289481564, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 42706.1, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.914, "characteristics.samples_per_second": 42706.1, "characteristics.samples_per_second.normalized_per_core": 42706.1, "characteristics.samples_per_second.normalized_per_processor": 42706.1, "ck_system": "A100-PCIex1_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-mobilenet", "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/A100-PCIex1_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": 1024, "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": 29040000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIex1_TRT", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 44000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "52e54bae4a18981d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 384495047042, "90.00 percentile latency (ns)": 692560411294, "95.00 percentile latency (ns)": 731014059168, "97.00 percentile latency (ns)": 746423062214, "99.00 percentile latency (ns)": 761817558694, "99.90 percentile latency (ns)": 768821779749, "Max latency (ns)": 769431749608, "Mean latency (ns)": 384486175893, "Min duration satisfied": "Yes", "Min latency (ns)": 153906184, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 823.465, "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": 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": 823.465, "characteristics.samples_per_second.normalized_per_core": 823.465, "characteristics.samples_per_second.normalized_per_processor": 823.465, "ck_system": "A100-PCIex1_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": 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/A100-PCIex1_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": 633600, "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-PCIex1_TRT", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "6fa0d9864445bf5f", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 315640883816, "90.00 percentile latency (ns)": 567855601266, "95.00 percentile latency (ns)": 599401824931, "97.00 percentile latency (ns)": 611971407937, "99.00 percentile latency (ns)": 624589422713, "99.90 percentile latency (ns)": 630263580871, "Max latency (ns)": 630853231460, "Mean latency (ns)": 315593952971, "Min duration satisfied": "Yes", "Min latency (ns)": 152733150, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1786.91, "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 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.mAP": 22.908, "characteristics.samples_per_second": 1786.91, "characteristics.samples_per_second.normalized_per_core": 1786.91, "characteristics.samples_per_second.normalized_per_processor": 1786.91, "ck_system": "Xavier_NX_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "informal_model": "ssd-mobilenet", "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/Xavier_NX_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "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": 1127280, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1708, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "b0afc56f64e74244", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 352891211183, "90.00 percentile latency (ns)": 641922211959, "95.00 percentile latency (ns)": 678069485343, "97.00 percentile latency (ns)": 692501199758, "99.00 percentile latency (ns)": 706938883963, "99.90 percentile latency (ns)": 713467709004, "Max latency (ns)": 714173906873, "Mean latency (ns)": 355189236057, "Min duration satisfied": "Yes", "Min latency (ns)": 45680463, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 36.9658, "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 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.mAP": 19.938, "characteristics.samples_per_second": 36.9658, "characteristics.samples_per_second.normalized_per_core": 36.9658, "characteristics.samples_per_second.normalized_per_processor": 36.9658, "ck_system": "Xavier_NX_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": "JetPack 4.6, TensorRT 8.0.1, 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 scenario", "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/Xavier_NX_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04", "other_hardware": "", "other_software_stack": "JetPack 4.6, TensorRT 8.0.1, CUDA 10.2, cuDNN 8.2.3, 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": 26400, "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/Xavier_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 40, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "7c3682220a466d92", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325729894235, "90.00 percentile latency (ns)": 586435988121, "95.00 percentile latency (ns)": 619048580587, "97.00 percentile latency (ns)": 632068107285, "99.00 percentile latency (ns)": 645127045426, "99.90 percentile latency (ns)": 650982995365, "Max latency (ns)": 651603870850, "Mean latency (ns)": 325727008096, "Min duration satisfied": "Yes", "Min latency (ns)": 75341722, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 24309.2, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.911, "characteristics.samples_per_second": 24309.2, "characteristics.samples_per_second.normalized_per_core": 24309.2, "characteristics.samples_per_second.normalized_per_processor": 24309.2, "ck_system": "A30x1_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-mobilenet", "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/A30x1_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": 1024, "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": 15840000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb", "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/A30x1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 24000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "d34b41850e644f1a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331582395503, "90.00 percentile latency (ns)": 596960384501, "95.00 percentile latency (ns)": 630144270534, "97.00 percentile latency (ns)": 643353899691, "99.00 percentile latency (ns)": 656627077689, "99.90 percentile latency (ns)": 662648293886, "Max latency (ns)": 663248991089, "Mean latency (ns)": 331603159175, "Min duration satisfied": "Yes", "Min latency (ns)": 69931232, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 467.698, "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": 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": 467.698, "characteristics.samples_per_second.normalized_per_core": 467.698, "characteristics.samples_per_second.normalized_per_processor": 467.698, "ck_system": "A30x1_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": 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/A30x1_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": 310200, "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/A30x1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 470, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "c9a910ee44a73a43", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 354843020908, "90.00 percentile latency (ns)": 637625223113, "95.00 percentile latency (ns)": 672967664843, "97.00 percentile latency (ns)": 687073921869, "99.00 percentile latency (ns)": 701225671457, "99.90 percentile latency (ns)": 707592524496, "Max latency (ns)": 708285721805, "Mean latency (ns)": 354513648340, "Min duration satisfied": "Yes", "Min latency (ns)": 184261335, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1425.7, "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 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.mAP": 22.881, "characteristics.power": 19.456252824858755, "characteristics.power.normalized_per_core": 19.456252824858755, "characteristics.power.normalized_per_processor": 19.456252824858755, "characteristics.samples_per_second": 1425.7, "characteristics.samples_per_second.normalized_per_core": 1425.7, "characteristics.samples_per_second.normalized_per_processor": 1425.7, "ck_system": "Xavier_NX_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": "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": "ssd-mobilenet", "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/Xavier_NX_TRT_MaxQ", "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": 1024, "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": 1009800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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_NX_TRT_MaxQ", "system_name": "NVIDIA Jetson Xavier NX (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1530, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "0a14fd789e884046", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 439506814983, "90.00 percentile latency (ns)": 790719211857, "95.00 percentile latency (ns)": 834907473867, "97.00 percentile latency (ns)": 852753945467, "99.00 percentile latency (ns)": 870327967175, "99.90 percentile latency (ns)": 878247325977, "Max latency (ns)": 879097769250, "Mean latency (ns)": 439593906013, "Min duration satisfied": "Yes", "Min latency (ns)": 58621120, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 30.0308, "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 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.mAP": 20.027, "characteristics.power": 19.882170648464186, "characteristics.power.normalized_per_core": 19.882170648464186, "characteristics.power.normalized_per_processor": 19.882170648464186, "characteristics.samples_per_second": 30.0308, "characteristics.samples_per_second.normalized_per_core": 30.0308, "characteristics.samples_per_second.normalized_per_processor": 30.0308, "ck_system": "Xavier_NX_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": "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": "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/Xavier_NX_TRT_MaxQ", "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": 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": 26400, "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/Xavier_NX_TRT_MaxQ", "system_name": "NVIDIA Jetson Xavier NX (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 40, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "4a8e821941462d57", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 391724164605, "90.00 percentile latency (ns)": 705906082493, "95.00 percentile latency (ns)": 745182700760, "97.00 percentile latency (ns)": 760891444178, "99.00 percentile latency (ns)": 776599086606, "99.90 percentile latency (ns)": 783664687606, "Max latency (ns)": 784427607773, "Mean latency (ns)": 391779759194, "Min duration satisfied": "Yes", "Min latency (ns)": 57486755, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 43078.5, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.912, "characteristics.samples_per_second": 43078.5, "characteristics.samples_per_second.normalized_per_core": 43078.5, "characteristics.samples_per_second.normalized_per_processor": 43078.5, "ck_system": "A100-PCIex1_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-mobilenet", "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/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": 1024, "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": 33792000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "64580b09388752f6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 385415210372, "90.00 percentile latency (ns)": 694671490728, "95.00 percentile latency (ns)": 733328738652, "97.00 percentile latency (ns)": 748795678724, "99.00 percentile latency (ns)": 764270180124, "99.90 percentile latency (ns)": 771224677136, "Max latency (ns)": 771928008071, "Mean latency (ns)": 385456020302, "Min duration satisfied": "Yes", "Min latency (ns)": 80297827, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 820.802, "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": 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": 820.802, "characteristics.samples_per_second.normalized_per_core": 820.802, "characteristics.samples_per_second.normalized_per_processor": 820.802, "ck_system": "A100-PCIex1_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": 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": 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": "", "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)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "24494bf3978b8548", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 333083472999, "90.00 percentile latency (ns)": 599141574770, "95.00 percentile latency (ns)": 632414235787, "97.00 percentile latency (ns)": 645646032514, "99.00 percentile latency (ns)": 658950912568, "99.90 percentile latency (ns)": 664919662648, "Max latency (ns)": 665536309273, "Mean latency (ns)": 333024046277, "Min duration satisfied": "Yes", "Min latency (ns)": 88890434, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2796.54, "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 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.mAP": 22.888, "characteristics.samples_per_second": 2796.54, "characteristics.samples_per_second.normalized_per_core": 2796.54, "characteristics.samples_per_second.normalized_per_processor": 2796.54, "ck_system": "AGX_Xavier_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": "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": "ssd-mobilenet", "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/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": 1024, "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": 1861200, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 2820, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "5a444fe28e596716", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326227684728, "90.00 percentile latency (ns)": 587061948392, "95.00 percentile latency (ns)": 619610841569, "97.00 percentile latency (ns)": 632613655996, "99.00 percentile latency (ns)": 645658933475, "99.90 percentile latency (ns)": 651540455036, "Max latency (ns)": 652172587854, "Mean latency (ns)": 326231956130, "Min duration satisfied": "Yes", "Min latency (ns)": 53627084, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 56.6721, "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 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.mAP": 20.052, "characteristics.samples_per_second": 56.6721, "characteristics.samples_per_second.normalized_per_core": 56.6721, "characteristics.samples_per_second.normalized_per_processor": 56.6721, "ck_system": "AGX_Xavier_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": "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": "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/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": 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": 36960, "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/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 56, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "0d8e75d4cd2abecb", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329812288262, "90.00 percentile latency (ns)": 593504990765, "95.00 percentile latency (ns)": 626469640956, "97.00 percentile latency (ns)": 639662706908, "99.00 percentile latency (ns)": 652850810565, "99.90 percentile latency (ns)": 658785479253, "Max latency (ns)": 659421985431, "Mean latency (ns)": 329801867518, "Min duration satisfied": "Yes", "Min latency (ns)": 296891252, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 51244.9, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.912, "characteristics.samples_per_second": 51244.9, "characteristics.samples_per_second.normalized_per_core": 51244.9, "characteristics.samples_per_second.normalized_per_processor": 51244.9, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_edge", "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-mobilenet", "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_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": 1024, "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": 33792000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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": "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)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "be84301973ea7a77", "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_edge", "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_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": 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_edge", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "ee15c58132264a2a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330603099498, "90.00 percentile latency (ns)": 595078124820, "95.00 percentile latency (ns)": 628124421973, "97.00 percentile latency (ns)": 641349900594, "99.00 percentile latency (ns)": 654575594630, "99.90 percentile latency (ns)": 660511163456, "Max latency (ns)": 661165990398, "Mean latency (ns)": 330601616270, "Min duration satisfied": "Yes", "Min latency (ns)": 38073324, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 6987.66, "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": 22.912, "characteristics.samples_per_second": 6987.66, "characteristics.samples_per_second.normalized_per_core": 6987.66, "characteristics.samples_per_second.normalized_per_processor": 6987.66, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.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-mobilenet", "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_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": 1024, "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": 4620000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_Triton", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 7000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "5c636c76204d07ba", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328957654658, "90.00 percentile latency (ns)": 591785174186, "95.00 percentile latency (ns)": 624845897800, "97.00 percentile latency (ns)": 637835515913, "99.00 percentile latency (ns)": 651060226094, "99.90 percentile latency (ns)": 656964088705, "Max latency (ns)": 657524935649, "Mean latency (ns)": 328883417379, "Min duration satisfied": "Yes", "Min latency (ns)": 238277682, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 135.508, "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": 135.508, "characteristics.samples_per_second.normalized_per_core": 135.508, "characteristics.samples_per_second.normalized_per_processor": 135.508, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.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": 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_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": 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_Triton", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 135, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "a774d595959b41c1", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 396431129993, "90.00 percentile latency (ns)": 713348679461, "95.00 percentile latency (ns)": 752935368040, "97.00 percentile latency (ns)": 768809580882, "99.00 percentile latency (ns)": 784638526007, "99.90 percentile latency (ns)": 791775717500, "Max latency (ns)": 792497380640, "Mean latency (ns)": 396397537108, "Min duration satisfied": "Yes", "Min latency (ns)": 227706853, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1648.96, "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 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.mAP": 22.919, "characteristics.power": 23.794607323232285, "characteristics.power.normalized_per_core": 23.794607323232285, "characteristics.power.normalized_per_processor": 23.794607323232285, "characteristics.samples_per_second": 1648.96, "characteristics.samples_per_second.normalized_per_core": 1648.96, "characteristics.samples_per_second.normalized_per_processor": 1648.96, "ck_system": "AGX_Xavier_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": "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": "ssd-mobilenet", "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/AGX_Xavier_TRT_MaxQ", "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": 1024, "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": 1306800, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/AGX_Xavier_TRT_MaxQ", "system_name": "NVIDIA Jetson AGX Xavier 32GB (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1980, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "b6bfb9ba8c48cdbc", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 427808882830, "90.00 percentile latency (ns)": 769551421047, "95.00 percentile latency (ns)": 812231432756, "97.00 percentile latency (ns)": 829267400707, "99.00 percentile latency (ns)": 846395201221, "99.90 percentile latency (ns)": 854090774081, "Max latency (ns)": 854908804653, "Mean latency (ns)": 427691886884, "Min duration satisfied": "Yes", "Min latency (ns)": 102388763, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 32.4245, "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 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.mAP": 20.015, "characteristics.power": 23.197250292397666, "characteristics.power.normalized_per_core": 23.197250292397666, "characteristics.power.normalized_per_processor": 23.197250292397666, "characteristics.samples_per_second": 32.4245, "characteristics.samples_per_second.normalized_per_core": 32.4245, "characteristics.samples_per_second.normalized_per_processor": 32.4245, "ck_system": "AGX_Xavier_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": "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": "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/AGX_Xavier_TRT_MaxQ", "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": 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": 27720, "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/AGX_Xavier_TRT_MaxQ", "system_name": "NVIDIA Jetson AGX Xavier 32GB (MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 42, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "a39b064a3db8cbe6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 333568672349, "90.00 percentile latency (ns)": 600251393813, "95.00 percentile latency (ns)": 633598142375, "97.00 percentile latency (ns)": 646932677581, "99.00 percentile latency (ns)": 660265764997, "99.90 percentile latency (ns)": 666282196045, "Max latency (ns)": 666931392392, "Mean latency (ns)": 333555369955, "Min duration satisfied": "Yes", "Min latency (ns)": 58278430, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 4809.49, "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.3gb 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": 22.911, "characteristics.samples_per_second": 4809.49, "characteristics.samples_per_second.normalized_per_core": 4809.49, "characteristics.samples_per_second.normalized_per_processor": 4809.49, "ck_system": "A30-MIG_1x1g.3gb_TRT", "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": "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": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "ssd-mobilenet", "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/A30-MIG_1x1g.3gb_TRT", "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 460.46, 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": 1024, "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": 3207600, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb", "status": "preview", "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/A30-MIG_1x1g.3gb_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 4860, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "58d4f10bbd5c6322", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 326738231721, "90.00 percentile latency (ns)": 588104347188, "95.00 percentile latency (ns)": 620774463186, "97.00 percentile latency (ns)": 633841945076, "99.00 percentile latency (ns)": 646910389955, "99.90 percentile latency (ns)": 652783889783, "Max latency (ns)": 653410811073, "Mean latency (ns)": 326722669985, "Min duration satisfied": "Yes", "Min latency (ns)": 34319334, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 121.21, "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.3gb 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": 121.21, "characteristics.samples_per_second.normalized_per_core": 121.21, "characteristics.samples_per_second.normalized_per_processor": 121.21, "ck_system": "A30-MIG_1x1g.3gb_TRT", "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": "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": "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": 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/A30-MIG_1x1g.3gb_TRT", "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 460.46, 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": 79200, "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": "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/A30-MIG_1x1g.3gb_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 120, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "b1b0e65ce44745bd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 338339494515, "90.00 percentile latency (ns)": 610115196473, "95.00 percentile latency (ns)": 644072886875, "97.00 percentile latency (ns)": 657675074306, "99.00 percentile latency (ns)": 671245892357, "99.90 percentile latency (ns)": 677358827595, "Max latency (ns)": 678029271187, "Mean latency (ns)": 338443537203, "Min duration satisfied": "Yes", "Min latency (ns)": 60972293, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 18494.8, "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 A10", "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": 22.911, "characteristics.samples_per_second": 18494.8, "characteristics.samples_per_second.normalized_per_core": 18494.8, "characteristics.samples_per_second.normalized_per_processor": 18494.8, "ck_system": "A10x1_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-mobilenet", "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/A10x1_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": 1024, "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": 12540000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb", "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/A10x1_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 19000, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "cd16ec4628bc2904", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 325985550156, "90.00 percentile latency (ns)": 587905967102, "95.00 percentile latency (ns)": 620682098142, "97.00 percentile latency (ns)": 633686626430, "99.00 percentile latency (ns)": 646791646164, "99.90 percentile latency (ns)": 652730032856, "Max latency (ns)": 653319114754, "Mean latency (ns)": 326177839451, "Min duration satisfied": "Yes", "Min latency (ns)": 102846778, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 313.17, "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": 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": 313.17, "characteristics.samples_per_second.normalized_per_core": 313.17, "characteristics.samples_per_second.normalized_per_processor": 313.17, "ck_system": "A10x1_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": 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/A10x1_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": 204600, "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/A10x1_TRT_Triton", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 310, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "fe5ab70d9b55929e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331751739284, "90.00 percentile latency (ns)": 597191013608, "95.00 percentile latency (ns)": 630381925487, "97.00 percentile latency (ns)": 643654636227, "99.00 percentile latency (ns)": 656924205807, "99.90 percentile latency (ns)": 662898970391, "Max latency (ns)": 663542641755, "Mean latency (ns)": 331746861948, "Min duration satisfied": "Yes", "Min latency (ns)": 43112163, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 50926.6, "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", "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": 22.912, "characteristics.samples_per_second": 50926.6, "characteristics.samples_per_second.normalized_per_core": 50926.6, "characteristics.samples_per_second.normalized_per_processor": 50926.6, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_Triton_edge", "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-mobilenet", "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_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": 1024, "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": 33792000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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": "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)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "1d700d0d4eb6aa8d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "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_edge", "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_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": 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_edge", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "f6a609747bcb7bbd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 339096548421, "90.00 percentile latency (ns)": 611710715925, "95.00 percentile latency (ns)": 645784075149, "97.00 percentile latency (ns)": 659441405893, "99.00 percentile latency (ns)": 673055729264, "99.90 percentile latency (ns)": 679193751329, "Max latency (ns)": 679865790341, "Mean latency (ns)": 339283698670, "Min duration satisfied": "Yes", "Min latency (ns)": 193180769, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 18444.8, "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 A10", "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": 22.911, "characteristics.samples_per_second": 18444.8, "characteristics.samples_per_second.normalized_per_core": 18444.8, "characteristics.samples_per_second.normalized_per_processor": 18444.8, "ck_system": "A10x1_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-mobilenet", "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/A10x1_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": 1024, "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": 12540000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb", "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/A10x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 19000, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "dbecc41e5cb6698b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 324154001563, "90.00 percentile latency (ns)": 583414950135, "95.00 percentile latency (ns)": 615852009330, "97.00 percentile latency (ns)": 628724663745, "99.00 percentile latency (ns)": 641697592431, "99.90 percentile latency (ns)": 647577616724, "Max latency (ns)": 648159951125, "Mean latency (ns)": 324142351043, "Min duration satisfied": "Yes", "Min latency (ns)": 103187850, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 315.663, "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": 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": 315.663, "characteristics.samples_per_second.normalized_per_core": 315.663, "characteristics.samples_per_second.normalized_per_processor": 315.663, "ck_system": "A10x1_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": 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/A10x1_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": 204600, "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/A10x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x A10, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 310, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "da9eed4ae4d141fb", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 328607652999, "90.00 percentile latency (ns)": 591254494150, "95.00 percentile latency (ns)": 624140607553, "97.00 percentile latency (ns)": 637310302883, "99.00 percentile latency (ns)": 650471869061, "99.90 percentile latency (ns)": 656351194627, "Max latency (ns)": 656973079089, "Mean latency (ns)": 328579861244, "Min duration satisfied": "Yes", "Min latency (ns)": 248024955, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 24110.6, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.911, "characteristics.samples_per_second": 24110.6, "characteristics.samples_per_second.normalized_per_core": 24110.6, "characteristics.samples_per_second.normalized_per_processor": 24110.6, "ck_system": "A30x1_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-mobilenet", "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/A30x1_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": 1024, "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": 15840000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb", "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/A30x1_TRT", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 24000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "beec1919e1b46d3a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 331691102090, "90.00 percentile latency (ns)": 597347749947, "95.00 percentile latency (ns)": 630523752694, "97.00 percentile latency (ns)": 643798190169, "99.00 percentile latency (ns)": 657088084059, "99.90 percentile latency (ns)": 663117656529, "Max latency (ns)": 663648566488, "Mean latency (ns)": 331724582043, "Min duration satisfied": "Yes", "Min latency (ns)": 138206427, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 467.416, "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": 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": 467.416, "characteristics.samples_per_second.normalized_per_core": 467.416, "characteristics.samples_per_second.normalized_per_processor": 467.416, "ck_system": "A30x1_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": 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/A30x1_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": 310200, "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/A30x1_TRT", "system_name": "Gigabyte G482-Z54 (1x A30, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 470, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "4965aee3f2d7df6a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327363336798, "90.00 percentile latency (ns)": 589258520713, "95.00 percentile latency (ns)": 621987723894, "97.00 percentile latency (ns)": 635067792114, "99.00 percentile latency (ns)": 648176938490, "99.90 percentile latency (ns)": 654089862418, "Max latency (ns)": 654720752112, "Mean latency (ns)": 327349779136, "Min duration satisfied": "Yes", "Min latency (ns)": 29016800, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 5090.72, "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.3gb 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": 22.912, "characteristics.samples_per_second": 5090.72, "characteristics.samples_per_second.normalized_per_core": 5090.72, "characteristics.samples_per_second.normalized_per_processor": 5090.72, "ck_system": "A30-MIG_1x1g.3gb_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, "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": 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-mobilenet", "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/A30-MIG_1x1g.3gb_TRT_Triton", "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 460.46, 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": 1024, "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": 3333000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb", "status": "preview", "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/A30-MIG_1x1g.3gb_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 5050, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "54d158efdd274a25", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329888456547, "90.00 percentile latency (ns)": 593770456913, "95.00 percentile latency (ns)": 626745055446, "97.00 percentile latency (ns)": 639955918938, "99.00 percentile latency (ns)": 653135589842, "99.90 percentile latency (ns)": 659084542171, "Max latency (ns)": 659724338405, "Mean latency (ns)": 329885880133, "Min duration satisfied": "Yes", "Min latency (ns)": 34349126, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 125.052, "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.3gb 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": 125.052, "characteristics.samples_per_second.normalized_per_core": 125.052, "characteristics.samples_per_second.normalized_per_processor": 125.052, "ck_system": "A30-MIG_1x1g.3gb_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, "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": 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", "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/A30-MIG_1x1g.3gb_TRT_Triton", "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 460.46, 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": 82500, "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": "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/A30-MIG_1x1g.3gb_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 125, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "8e52c94e84b56ef2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 320121140113, "90.00 percentile latency (ns)": 576177264969, "95.00 percentile latency (ns)": 608174084021, "97.00 percentile latency (ns)": 620979501867, "99.00 percentile latency (ns)": 633785277945, "99.90 percentile latency (ns)": 639531967669, "Max latency (ns)": 640166088494, "Mean latency (ns)": 320116292933, "Min duration satisfied": "Yes", "Min latency (ns)": 68808748, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7216.88, "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": 22.912, "characteristics.samples_per_second": 7216.88, "characteristics.samples_per_second.normalized_per_core": 7216.88, "characteristics.samples_per_second.normalized_per_processor": 7216.88, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_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-mobilenet", "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", "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": 1024, "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": 4620000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 7000, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "910d4f4a4143bd9c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329053480267, "90.00 percentile latency (ns)": 591532585058, "95.00 percentile latency (ns)": 624581098696, "97.00 percentile latency (ns)": 637799464901, "99.00 percentile latency (ns)": 651018363544, "99.90 percentile latency (ns)": 656684157936, "Max latency (ns)": 657245889920, "Mean latency (ns)": 328862852699, "Min duration satisfied": "Yes", "Min latency (ns)": 472958598, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 135.566, "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": 135.566, "characteristics.samples_per_second.normalized_per_core": 135.566, "characteristics.samples_per_second.normalized_per_processor": 135.566, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_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": 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", "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", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 135, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "bdc84fd604a41628", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 391706104808, "90.00 percentile latency (ns)": 705735172423, "95.00 percentile latency (ns)": 744987178116, "97.00 percentile latency (ns)": 760687975405, "99.00 percentile latency (ns)": 776392042779, "99.90 percentile latency (ns)": 783458272131, "Max latency (ns)": 784220287059, "Mean latency (ns)": 391777179123, "Min duration satisfied": "Yes", "Min latency (ns)": 310751983, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 43089.9, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 22.912, "characteristics.samples_per_second": 43089.9, "characteristics.samples_per_second.normalized_per_core": 43089.9, "characteristics.samples_per_second.normalized_per_processor": 43089.9, "ck_system": "A100-PCIex1_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-mobilenet", "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/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": 1024, "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": 33792000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIex1_TRT", "system_name": "Gigabyte G482-Z54 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "ce1f0a346ad64621", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 382631936069, "90.00 percentile latency (ns)": 689569571535, "95.00 percentile latency (ns)": 727872884804, "97.00 percentile latency (ns)": 743219268216, "99.00 percentile latency (ns)": 758567209957, "99.90 percentile latency (ns)": 765526732427, "Max latency (ns)": 766146768066, "Mean latency (ns)": 382662694919, "Min duration satisfied": "Yes", "Min latency (ns)": 160615253, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 826.996, "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": 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": 826.996, "characteristics.samples_per_second.normalized_per_core": 826.996, "characteristics.samples_per_second.normalized_per_processor": 826.996, "ck_system": "A100-PCIex1_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": 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": 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": "", "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)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "c4c9dba5dc9b0b5e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 366207332462, "90.00 percentile latency (ns)": 659083431787, "95.00 percentile latency (ns)": 695723412179, "97.00 percentile latency (ns)": 710363140998, "99.00 percentile latency (ns)": 724998331603, "99.90 percentile latency (ns)": 731595387844, "Max latency (ns)": 732319866638, "Mean latency (ns)": 366207127554, "Min duration satisfied": "Yes", "Min latency (ns)": 81505397, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1539.33, "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 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.mAP": 22.911, "characteristics.samples_per_second": 1539.33, "characteristics.samples_per_second.normalized_per_core": 1539.33, "characteristics.samples_per_second.normalized_per_processor": 1539.33, "ck_system": "Xavier_NX_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": "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": "ssd-mobilenet", "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/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": 1024, "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": 1127280, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1708, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "a0764dddbf3329cd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 367775318277, "90.00 percentile latency (ns)": 662061241768, "95.00 percentile latency (ns)": 698850750434, "97.00 percentile latency (ns)": 713620016585, "99.00 percentile latency (ns)": 728305507159, "99.90 percentile latency (ns)": 734946646600, "Max latency (ns)": 735638972722, "Mean latency (ns)": 367813073673, "Min duration satisfied": "Yes", "Min latency (ns)": 48060448, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 35.8872, "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 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.mAP": 20.034, "characteristics.samples_per_second": 35.8872, "characteristics.samples_per_second.normalized_per_core": 35.8872, "characteristics.samples_per_second.normalized_per_processor": 35.8872, "ck_system": "Xavier_NX_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": "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": "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/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": 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": 26400, "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/Xavier_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 40, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "7e49d6a9939e2f50", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32940505241, "90.00 percentile latency (ns)": 59401325080, "95.00 percentile latency (ns)": 62706615985, "97.00 percentile latency (ns)": 64030389187, "99.00 percentile latency (ns)": 65353656801, "99.90 percentile latency (ns)": 65934338504, "Max latency (ns)": 65994089856, "Mean latency (ns)": 32955476667, "Min duration satisfied": "Yes", "Min latency (ns)": 45414436, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 51204.6, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.911, "characteristics.samples_per_second": 51204.6, "characteristics.samples_per_second.normalized_per_core": 51204.6, "characteristics.samples_per_second.normalized_per_processor": 51204.6, "ck_system": "DGX-A100_A100-SXM4x1_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-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 3379200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/DGX-A100_A100-SXM4x1_TRT", "system_name": "NVIDIA DGX-A100 (1x A100-SXM4, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "803cb4df29d71fa2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32314222181, "90.00 percentile latency (ns)": 58211153341, "95.00 percentile latency (ns)": 61471122534, "97.00 percentile latency (ns)": 62780363434, "99.00 percentile latency (ns)": 64086744368, "99.90 percentile latency (ns)": 64610709623, "Max latency (ns)": 64610709623, "Mean latency (ns)": 32327192445, "Min duration satisfied": "Yes", "Min latency (ns)": 127367628, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 980.642, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 980.642, "characteristics.samples_per_second.normalized_per_core": 980.642, "characteristics.samples_per_second.normalized_per_processor": 980.642, "ck_system": "DGX-A100_A100-SXM4x1_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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 63360, "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-SXM4x1_TRT", "system_name": "NVIDIA DGX-A100 (1x A100-SXM4, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "e4c7fb833da5b57c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 246975819656, "90.00 percentile latency (ns)": 444547456127, "95.00 percentile latency (ns)": 469243441758, "97.00 percentile latency (ns)": 479122028505, "99.00 percentile latency (ns)": 489000375122, "99.90 percentile latency (ns)": 493434442012, "Max latency (ns)": 493920600627, "Mean latency (ns)": 246968292539, "Min duration satisfied": "Yes", "Min latency (ns)": 21505345, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 6841.59, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "5GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4 (1x1g.5gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 6841.59, "characteristics.samples_per_second.normalized_per_core": 6841.59, "characteristics.samples_per_second.normalized_per_processor": 6841.59, "ck_system": "DGX-A100_A100-SXM4x1-MIG_1x1g.5gb_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-mobilenet", "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": 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-MIG_1x1g.5gb_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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 3379200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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": "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-MIG_1x1g.5gb_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "3f4a9a99334fa8fd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 235352001471, "90.00 percentile latency (ns)": 423438323996, "95.00 percentile latency (ns)": 446950321532, "97.00 percentile latency (ns)": 456212966577, "99.00 percentile latency (ns)": 465713029746, "99.90 percentile latency (ns)": 469987761522, "Max latency (ns)": 470225197550, "Mean latency (ns)": 235233368260, "Min duration satisfied": "Yes", "Min latency (ns)": 243672825, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 134.744, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "5GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4 (1x1g.5gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 134.744, "characteristics.samples_per_second.normalized_per_core": 134.744, "characteristics.samples_per_second.normalized_per_processor": 134.744, "ck_system": "DGX-A100_A100-SXM4x1-MIG_1x1g.5gb_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": 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-MIG_1x1g.5gb_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": 63360, "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-SXM4x1-MIG_1x1g.5gb_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "a260974acf84511a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 38159957551, "90.00 percentile latency (ns)": 68883803023, "95.00 percentile latency (ns)": 72736581079, "97.00 percentile latency (ns)": 74276250610, "99.00 percentile latency (ns)": 75815974643, "99.90 percentile latency (ns)": 76494958622, "Max latency (ns)": 76566058795, "Mean latency (ns)": 38200838526, "Min duration satisfied": "Yes", "Min latency (ns)": 38637244, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 44134.4, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 44134.4, "characteristics.samples_per_second.normalized_per_core": 44134.4, "characteristics.samples_per_second.normalized_per_processor": 44134.4, "ck_system": "A100-PCIex1_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, "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-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 3379200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z52 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "b244b468ebfc124f", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 37076590237, "90.00 percentile latency (ns)": 66937067297, "95.00 percentile latency (ns)": 70642252526, "97.00 percentile latency (ns)": 72156533438, "99.00 percentile latency (ns)": 73669101125, "99.90 percentile latency (ns)": 74350045822, "Max latency (ns)": 74350194289, "Mean latency (ns)": 37099491284, "Min duration satisfied": "Yes", "Min latency (ns)": 80995906, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 852.183, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 852.183, "characteristics.samples_per_second.normalized_per_core": 852.183, "characteristics.samples_per_second.normalized_per_processor": 852.183, "ck_system": "A100-PCIex1_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, "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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 63360, "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-PCIex1_TRT_Triton", "system_name": "Gigabyte G482-Z52 (1x A100-PCIe, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "314b9a83d14d9688", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32635988906, "90.00 percentile latency (ns)": 58717392967, "95.00 percentile latency (ns)": 61927462737, "97.00 percentile latency (ns)": 63246952778, "99.00 percentile latency (ns)": 64540881365, "99.90 percentile latency (ns)": 65136281492, "Max latency (ns)": 65146754278, "Mean latency (ns)": 32630411181, "Min duration satisfied": "Yes", "Min latency (ns)": 68231028, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 2532.74, "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 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.mAP": 22.927, "characteristics.samples_per_second": 2532.74, "characteristics.samples_per_second.normalized_per_core": 2532.74, "characteristics.samples_per_second.normalized_per_processor": 2532.74, "ck_system": "AGX_Xavier_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": "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": "ssd-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 165000, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 2500, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "feaaf8370cebfa2c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 241129794292, "90.00 percentile latency (ns)": 433998013294, "95.00 percentile latency (ns)": 458064039907, "97.00 percentile latency (ns)": 467683455185, "99.00 percentile latency (ns)": 477300353795, "99.90 percentile latency (ns)": 481594009269, "Max latency (ns)": 482111755230, "Mean latency (ns)": 241112094024, "Min duration satisfied": "Yes", "Min latency (ns)": 58421532, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 50.9757, "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 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.mAP": 20.048, "characteristics.samples_per_second": 50.9757, "characteristics.samples_per_second.normalized_per_core": 50.9757, "characteristics.samples_per_second.normalized_per_processor": 50.9757, "ck_system": "AGX_Xavier_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": "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": "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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "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/AGX_Xavier_TRT", "system_name": "NVIDIA Jetson AGX Xavier 32GB (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 50, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "486d9233e5c84280", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 33226550855, "90.00 percentile latency (ns)": 59883168626, "95.00 percentile latency (ns)": 63214459967, "97.00 percentile latency (ns)": 64546363210, "99.00 percentile latency (ns)": 65878751214, "99.90 percentile latency (ns)": 66464693736, "Max latency (ns)": 66524750791, "Mean latency (ns)": 33227949918, "Min duration satisfied": "Yes", "Min latency (ns)": 36773145, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 50796.1, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.911, "characteristics.samples_per_second": 50796.1, "characteristics.samples_per_second.normalized_per_core": 50796.1, "characteristics.samples_per_second.normalized_per_processor": 50796.1, "ck_system": "DGX-A100_A100-SXM4x1_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-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 3379200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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": "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)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "24493db71fff6233", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32352776178, "90.00 percentile latency (ns)": 58296057018, "95.00 percentile latency (ns)": 61506415463, "97.00 percentile latency (ns)": 62816635419, "99.00 percentile latency (ns)": 64126443884, "99.90 percentile latency (ns)": 64715938930, "Max latency (ns)": 64716002299, "Mean latency (ns)": 32340127332, "Min duration satisfied": "Yes", "Min latency (ns)": 74669698, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 979.047, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 979.047, "characteristics.samples_per_second.normalized_per_core": 979.047, "characteristics.samples_per_second.normalized_per_processor": 979.047, "ck_system": "DGX-A100_A100-SXM4x1_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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 63360, "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-SXM4x1_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "ec4bf1295e5cfdb8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 243509115494, "90.00 percentile latency (ns)": 438289812217, "95.00 percentile latency (ns)": 462640354053, "97.00 percentile latency (ns)": 472379832844, "99.00 percentile latency (ns)": 482117362879, "99.90 percentile latency (ns)": 486487510384, "Max latency (ns)": 486966938105, "Mean latency (ns)": 243502986083, "Min duration satisfied": "Yes", "Min latency (ns)": 44819993, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 6939.28, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "5GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4 (1x1g.5gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 6939.28, "characteristics.samples_per_second.normalized_per_core": 6939.28, "characteristics.samples_per_second.normalized_per_processor": 6939.28, "ck_system": "DGX-A100_A100-SXM4x1-MIG_1x1g.5gb_TRT", "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-mobilenet", "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": 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-MIG_1x1g.5gb_TRT", "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; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 3379200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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": "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-MIG_1x1g.5gb_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "72aa455a83804066", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 236133763302, "90.00 percentile latency (ns)": 424663065387, "95.00 percentile latency (ns)": 447995939504, "97.00 percentile latency (ns)": 457520884231, "99.00 percentile latency (ns)": 467044784388, "99.90 percentile latency (ns)": 471328575174, "Max latency (ns)": 471328677225, "Mean latency (ns)": 235899112443, "Min duration satisfied": "Yes", "Min latency (ns)": 476985660, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 134.428, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "5GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4 (1x1g.5gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 134.428, "characteristics.samples_per_second.normalized_per_core": 134.428, "characteristics.samples_per_second.normalized_per_processor": 134.428, "ck_system": "DGX-A100_A100-SXM4x1-MIG_1x1g.5gb_TRT", "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": 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-MIG_1x1g.5gb_TRT", "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; 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": 63360, "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-SXM4x1-MIG_1x1g.5gb_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "dc43c19a04ac584c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 30790982911, "90.00 percentile latency (ns)": 55629721138, "95.00 percentile latency (ns)": 58736883555, "97.00 percentile latency (ns)": 59986848351, "99.00 percentile latency (ns)": 61238144596, "99.90 percentile latency (ns)": 61808104546, "Max latency (ns)": 61859688365, "Mean latency (ns)": 30836752049, "Min duration satisfied": "Yes", "Min latency (ns)": 15597586, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 7962.5, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.911, "characteristics.samples_per_second": 7962.5, "characteristics.samples_per_second.normalized_per_core": 7962.5, "characteristics.samples_per_second.normalized_per_processor": 7962.5, "ck_system": "T4x1_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-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 492558, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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": "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)": 0, "target_qps": 7463, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "545829d6c1303359", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 90632571630, "90.00 percentile latency (ns)": 163681644063, "95.00 percentile latency (ns)": 172791100349, "97.00 percentile latency (ns)": 176451364415, "99.00 percentile latency (ns)": 180113699898, "99.90 percentile latency (ns)": 181720957417, "Max latency (ns)": 181899967289, "Mean latency (ns)": 90715097074, "Min duration satisfied": "Yes", "Min latency (ns)": 77760517, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 135.107, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 135.107, "characteristics.samples_per_second.normalized_per_core": 135.107, "characteristics.samples_per_second.normalized_per_processor": 135.107, "ck_system": "T4x1_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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "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/T4x1_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 140, "task": "object detection", "task2": "object detection", "total_cores": 240, "uid": "9ab51c6b644458da", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 30806178314, "90.00 percentile latency (ns)": 55652163106, "95.00 percentile latency (ns)": 58762899027, "97.00 percentile latency (ns)": 60015108996, "99.00 percentile latency (ns)": 61263171636, "99.90 percentile latency (ns)": 61830186645, "Max latency (ns)": 61881013571, "Mean latency (ns)": 30851495920, "Min duration satisfied": "Yes", "Min latency (ns)": 21097154, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 7959.76, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.911, "characteristics.samples_per_second": 7959.76, "characteristics.samples_per_second.normalized_per_core": 7959.76, "characteristics.samples_per_second.normalized_per_processor": 7959.76, "ck_system": "T4x1_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-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 492558, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/T4x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x T4, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 7463, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "667393c48d99e9fb", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 90418696500, "90.00 percentile latency (ns)": 163260092674, "95.00 percentile latency (ns)": 172345382738, "97.00 percentile latency (ns)": 176001089966, "99.00 percentile latency (ns)": 179656445591, "99.90 percentile latency (ns)": 181259601091, "Max latency (ns)": 181436960404, "Mean latency (ns)": 90494673134, "Min duration satisfied": "Yes", "Min latency (ns)": 78857549, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 135.452, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 135.452, "characteristics.samples_per_second.normalized_per_core": 135.452, "characteristics.samples_per_second.normalized_per_processor": 135.452, "ck_system": "T4x1_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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "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/T4x1_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (1x T4, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 140, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "ce20b104c9ca9f45", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 37953858170, "90.00 percentile latency (ns)": 68519648582, "95.00 percentile latency (ns)": 72349975779, "97.00 percentile latency (ns)": 73881772891, "99.00 percentile latency (ns)": 75414952449, "99.90 percentile latency (ns)": 76088255188, "Max latency (ns)": 76157961186, "Mean latency (ns)": 37998567222, "Min duration satisfied": "Yes", "Min latency (ns)": 58544734, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 44370.9, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.912, "characteristics.samples_per_second": 44370.9, "characteristics.samples_per_second.normalized_per_core": 44370.9, "characteristics.samples_per_second.normalized_per_processor": 44370.9, "ck_system": "A100-PCIex1_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-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 3379200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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/A100-PCIex1_TRT", "system_name": "Gigabyte G482-Z52 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 51200, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "1e4af37ee346fdad", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 37093386567, "90.00 percentile latency (ns)": 66947898800, "95.00 percentile latency (ns)": 70718393308, "97.00 percentile latency (ns)": 72232114594, "99.00 percentile latency (ns)": 73745593774, "99.90 percentile latency (ns)": 74349935631, "Max latency (ns)": 74349935631, "Mean latency (ns)": 37144255062, "Min duration satisfied": "Yes", "Min latency (ns)": 153082687, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 852.186, "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.111, "characteristics.samples_per_second": 852.186, "characteristics.samples_per_second.normalized_per_core": 852.186, "characteristics.samples_per_second.normalized_per_processor": 852.186, "ck_system": "A100-PCIex1_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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 63360, "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-PCIex1_TRT", "system_name": "Gigabyte G482-Z52 (1x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 960, "task": "object detection", "task2": "object detection", "total_cores": 128, "uid": "94a398c276ccffe6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 44109344971, "90.00 percentile latency (ns)": 80008738935, "95.00 percentile latency (ns)": 84536155681, "97.00 percentile latency (ns)": 86314494388, "99.00 percentile latency (ns)": 88088135523, "99.90 percentile latency (ns)": 88936174132, "Max latency (ns)": 88988069454, "Mean latency (ns)": 44244324607, "Min duration satisfied": "Yes", "Min latency (ns)": 70381304, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 1266.78, "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 NX", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 22.95, "characteristics.samples_per_second": 1266.78, "characteristics.samples_per_second.normalized_per_core": 1266.78, "characteristics.samples_per_second.normalized_per_processor": 1266.78, "ck_system": "Xavier_NX_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": "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": "ssd-mobilenet", "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": 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": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 112728, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "ssd_mobilenet_v1_coco_2018_01_28/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_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1708, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "05a71d490e9011f8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 456750191271, "90.00 percentile latency (ns)": 825457945143, "95.00 percentile latency (ns)": 871543148656, "97.00 percentile latency (ns)": 889993572562, "99.00 percentile latency (ns)": 908449226139, "99.90 percentile latency (ns)": 916707342407, "Max latency (ns)": 917614884419, "Mean latency (ns)": 456972593203, "Min duration satisfied": "Yes", "Min latency (ns)": 58307272, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 26.7825, "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 NX", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.mAP": 20.027, "characteristics.samples_per_second": 26.7825, "characteristics.samples_per_second.normalized_per_core": 26.7825, "characteristics.samples_per_second.normalized_per_processor": 26.7825, "ck_system": "Xavier_NX_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": "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": "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": 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": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 24576, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "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_NX_TRT", "system_name": "NVIDIA Jetson Xavier NX (TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 44, "task": "object detection", "task2": "object detection", "total_cores": 6, "uid": "b43bf9e00dbb1a28", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]