[ { "50.00 percentile latency (ns)": 659819053963, "90.00 percentile latency (ns)": 1188567633870, "95.00 percentile latency (ns)": 1254684407272, "97.00 percentile latency (ns)": 1281121832870, "99.00 percentile latency (ns)": 1307562759096, "99.90 percentile latency (ns)": 1319474734085, "Max latency (ns)": 1320732459703, "Mean latency (ns)": 659902820573, "Min duration satisfied": "Yes", "Min latency (ns)": 177499436, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3897.84, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 20.117, "characteristics.samples_per_second": 3897.84, "characteristics.samples_per_second.normalized_per_core": 974.46, "characteristics.samples_per_second.normalized_per_processor": 974.46, "ck_system": "HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.4", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/HPE/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/HPE/results/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.4, 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": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 5148000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "status": "available", "submitter": "HPE", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/HPE", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM-80GBx4_TRT", "system_name": "HPE ProLiant XL645d Gen10 Plus (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7800, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "e6a2b06dd086fb35", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 661873609587, "90.00 percentile latency (ns)": 1192109462338, "95.00 percentile latency (ns)": 1258399920364, "97.00 percentile latency (ns)": 1284917575000, "99.00 percentile latency (ns)": 1311436833962, "99.90 percentile latency (ns)": 1323376900172, "Max latency (ns)": 1324681144497, "Mean latency (ns)": 661934521094, "Min duration satisfied": "Yes", "Min latency (ns)": 89620879, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 3886.22, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 20.117, "characteristics.samples_per_second": 3886.22, "characteristics.samples_per_second.normalized_per_core": 971.555, "characteristics.samples_per_second.normalized_per_processor": 971.555, "ck_system": "HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 1, "host_storage_capacity": "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": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/HPE/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/HPE/results/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM-80GBx4_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, 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": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 5148000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet34-ssd1200.pytorch", "status": "available", "submitter": "HPE", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/HPE", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM-80GBx4_TRT_Triton", "system_name": "HPE ProLiant XL645d Gen10 Plus (4x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7800, "task": "object detection", "task2": "object detection", "total_cores": 64, "uid": "6ffe4aff82e4236b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 337341718248, "90.00 percentile latency (ns)": 608333260838, "95.00 percentile latency (ns)": 642289766609, "97.00 percentile latency (ns)": 655880567955, "99.00 percentile latency (ns)": 669474915307, "99.90 percentile latency (ns)": 675589803288, "Max latency (ns)": 676268446902, "Mean latency (ns)": 337527236787, "Min duration satisfied": "Yes", "Min latency (ns)": 24489955, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 3747.62, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM4-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 20.111, "characteristics.samples_per_second": 3747.62, "characteristics.samples_per_second.normalized_per_core": 936.905, "characteristics.samples_per_second.normalized_per_processor": 936.905, "ck_system": "HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM4-40GBx4_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": "512 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": 1, "host_storage_capacity": "3.2 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/HPE/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/HPE/results/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM4-40GBx4_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 460.27.04, 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": 2534400, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "status": "available", "submitter": "HPE", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/HPE", "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/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM4-40GBx4_TRT_Triton", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 3840, "task": "object detection", "task2": "object detection", "total_cores": 120, "uid": "38666757607197e1", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 327211089314, "90.00 percentile latency (ns)": 589954705232, "95.00 percentile latency (ns)": 622815679497, "97.00 percentile latency (ns)": 635954733666, "99.00 percentile latency (ns)": 649101164920, "99.90 percentile latency (ns)": 655008854824, "Max latency (ns)": 655673930806, "Mean latency (ns)": 327393278813, "Min duration satisfied": "Yes", "Min latency (ns)": 50741775, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 3865.34, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM4-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.mAP": 20.111, "characteristics.samples_per_second": 3865.34, "characteristics.samples_per_second.normalized_per_core": 966.335, "characteristics.samples_per_second.normalized_per_processor": 966.335, "ck_system": "HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM4-40GBx4_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": "512 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": 1, "host_storage_capacity": "3.2 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "ssd-resnet34", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/HPE/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/HPE/results/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM4-40GBx4_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 460.27.04, 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": 2534400, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "resnet34-ssd1200.pytorch", "status": "available", "submitter": "HPE", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/HPE", "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/HPE-ProLiant-XL645d-Gen10-Plus_A100-SXM4-40GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 3840, "task": "object detection", "task2": "object detection", "total_cores": 120, "uid": "43b0853801c67205", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]