[ { "50.00 percentile latency (ns)": 317311239853, "90.00 percentile latency (ns)": 571274294404, "95.00 percentile latency (ns)": 603025120606, "97.00 percentile latency (ns)": 615727146277, "99.00 percentile latency (ns)": 628416099406, "99.90 percentile latency (ns)": 634158652309, "Max latency (ns)": 634769575442, "Mean latency (ns)": 26215548144, "Min duration satisfied": "Yes", "Min latency (ns)": 467035319, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 99815.7, "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": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 99815.7, "characteristics.samples_per_second.normalized_per_core": 33271.9, "characteristics.samples_per_second.normalized_per_processor": 33271.9, "characteristics.total": 50000, "ck_system": "R7525_A100-PCIE-40GBx3_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 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": 32, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R7525_A100-PCIE-40GBx3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 63360000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A100-PCIE-40GBx3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A100-PCIE-40GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 96000, "task": "image classification", "task2": "image classification", "total_cores": 64, "uid": "e770c27028d4a639", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 332108531704, "90.00 percentile latency (ns)": 595891165856, "95.00 percentile latency (ns)": 628820634214, "97.00 percentile latency (ns)": 641986238545, "99.00 percentile latency (ns)": 655149389589, "99.90 percentile latency (ns)": 661084602744, "Max latency (ns)": 661733430092, "Mean latency (ns)": 142027100678, "Min duration satisfied": "Yes", "Min latency (ns)": 1012781116, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 146878, "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-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 146878, "characteristics.samples_per_second.normalized_per_core": 36719.5, "characteristics.samples_per_second.normalized_per_processor": 36719.5, "characteristics.total": 50000, "ck_system": "R750xa_A100-PCIE-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 97194240, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 147264, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "a21b22ea89718c24", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 373933311592, "90.00 percentile latency (ns)": 671266044911, "95.00 percentile latency (ns)": 708436873856, "97.00 percentile latency (ns)": 723289694064, "99.00 percentile latency (ns)": 738167385732, "99.90 percentile latency (ns)": 748109375489, "Max latency (ns)": 751493166668, "Mean latency (ns)": 14690321308, "Min duration satisfied": "Yes", "Min latency (ns)": 1932054260, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 136656, "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-PCIE-80GB (7x1g.10gb MIG)", "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.accuracy": 76.052, "characteristics.good": 38026, "characteristics.samples_per_second": 136656, "characteristics.samples_per_second.normalized_per_core": 34164.0, "characteristics.samples_per_second.normalized_per_processor": 34164.0, "characteristics.total": 50000, "ck_system": "R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "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": 2048, "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": 102696000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 155600, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "dc4dc45cba54eb95", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 361896447137, "90.00 percentile latency (ns)": 652679189888, "95.00 percentile latency (ns)": 689006322256, "97.00 percentile latency (ns)": 703507915870, "99.00 percentile latency (ns)": 718083754952, "99.90 percentile latency (ns)": 724629421387, "Max latency (ns)": 725309403126, "Mean latency (ns)": 362113505659, "Min duration satisfied": "Yes", "Min latency (ns)": 323863186, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 27298.7, "Scenario": "offline", "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.accuracy": 76.03, "characteristics.good": 38015, "characteristics.power": 566.118206896551, "characteristics.power.normalized_per_core": 283.0591034482755, "characteristics.power.normalized_per_processor": 283.0591034482755, "characteristics.samples_per_second": 27298.7, "characteristics.samples_per_second.normalized_per_core": 13649.35, "characteristics.samples_per_second.normalized_per_processor": 13649.35, "characteristics.total": 50000, "ck_system": "XE2420_A10x2_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "resnet50-v1.5", "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_cooling_type": "air", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6252 CPU @ 2.10GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "5.00.00.00", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE2420_A10x2_TRT_MaxQ", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_settings": "closed/Dell/power/XE2420_A10x2_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 19800000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_A10x2_TRT_MaxQ", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "image classification", "task2": "image classification", "total_cores": 48, "uid": "448f4b4db44cdc1e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 351265289474, "90.00 percentile latency (ns)": 630764551698, "95.00 percentile latency (ns)": 665715498913, "97.00 percentile latency (ns)": 679688853138, "99.00 percentile latency (ns)": 693674209075, "99.90 percentile latency (ns)": 699962200420, "Max latency (ns)": 700648757074, "Mean latency (ns)": 12517035476, "Min duration satisfied": "Yes", "Min latency (ns)": 2362241281, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 310855, "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-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 310855, "characteristics.samples_per_second.normalized_per_core": 31085.5, "characteristics.samples_per_second.normalized_per_processor": 31085.5, "characteristics.total": 50000, "ck_system": "DSS8440_A100-PCIE-80GBx10_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": " 768GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 10, "normalize_processors": 10, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/DSS8440_A100-PCIE-80GBx10_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 217800000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A100-PCIE-80GBx10_TRT", "system_name": "Dell EMC DSS 8440 (10x NVIDIA A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 330000, "task": "image classification", "task2": "image classification", "total_cores": 48, "uid": "d707561b454aac93", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330287892330, "90.00 percentile latency (ns)": 593860560288, "95.00 percentile latency (ns)": 626787744839, "97.00 percentile latency (ns)": 639964953555, "99.00 percentile latency (ns)": 653120169577, "99.90 percentile latency (ns)": 659065359471, "Max latency (ns)": 659718836326, "Mean latency (ns)": 140312980079, "Min duration satisfied": "Yes", "Min latency (ns)": 73907360, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 147327, "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-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 147327, "characteristics.samples_per_second.normalized_per_core": 36831.75, "characteristics.samples_per_second.normalized_per_processor": 36831.75, "characteristics.total": 50000, "ck_system": "R750xa_A100-PCIE-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "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": 2048, "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": 97194240, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT_Triton", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 147264, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "c8a75b184727bda3", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 322049055323, "90.00 percentile latency (ns)": 579310364329, "95.00 percentile latency (ns)": 611528521896, "97.00 percentile latency (ns)": 624325671489, "99.00 percentile latency (ns)": 637213027705, "99.90 percentile latency (ns)": 642981432277, "Max latency (ns)": 643656512910, "Mean latency (ns)": 119495276679, "Min duration satisfied": "Yes", "Min latency (ns)": 1119709144, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 141504, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 141504, "characteristics.samples_per_second.normalized_per_core": 17688.0, "characteristics.samples_per_second.normalized_per_processor": 17688.0, "characteristics.total": 50000, "ck_system": "DSS8440_A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/DSS8440_A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 91080000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 138000, "task": "image classification", "task2": "image classification", "total_cores": 96, "uid": "35b9bee2a9e6bcfc", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 333954478711, "90.00 percentile latency (ns)": 601437892372, "95.00 percentile latency (ns)": 634808459066, "97.00 percentile latency (ns)": 648167541858, "99.00 percentile latency (ns)": 661641503089, "99.90 percentile latency (ns)": 667625484934, "Max latency (ns)": 668222771438, "Mean latency (ns)": 333977111249, "Min duration satisfied": "Yes", "Min latency (ns)": 363152058, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 53928.1, "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": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 53928.1, "characteristics.samples_per_second.normalized_per_core": 17976.033333333333, "characteristics.samples_per_second.normalized_per_processor": 17976.033333333333, "characteristics.total": 50000, "ck_system": "R7525_A30x3_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7763", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R7525_A30x3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.4.2105", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "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": 36036000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A30x3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 54600, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "f14ee3950056a1d5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 807675394454, "90.00 percentile latency (ns)": 807675394454, "95.00 percentile latency (ns)": 807675394454, "97.00 percentile latency (ns)": 807675394454, "99.00 percentile latency (ns)": 807675394454, "99.90 percentile latency (ns)": 807675394454, "Max latency (ns)": 807675394454, "Mean latency (ns)": 807675394454, "Min duration satisfied": "Yes", "Min latency (ns)": 807675394454, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "SUT\u0000", "Samples per second": 2860.06, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 76.292, "characteristics.good": 38146, "characteristics.samples_per_second": 2860.06, "characteristics.samples_per_second.normalized_per_core": 51.0725, "characteristics.samples_per_second.normalized_per_processor": 1430.03, "characteristics.total": 50000, "ck_system": "R750_Intel6330_openvino", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "OpenVINO", "host_memory_capacity": "2.16TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "2.0GHz", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6330 CPU @ 2.0GHz", "host_processors_per_node": 2, "host_storage_capacity": "1.5TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 56, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750_Intel6330_openvino", "number_of_nodes": 1, "operating_system": "Ubuntu 20.10 (GNU/Linux 5.8.0-45-generic x86_64)", "other_software_stack": "MKL-DNN; URL: github.com/intel/mkl-dnn", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "no", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 2310000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_int8.xml", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750_Intel6330_openvino", "system_name": "Dell EMC PowerEdge R750 ICX-6330(2S 28C)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 3500, "task": "image classification", "task2": "image classification", "total_cores": 56, "uid": "249e8abffb4f5413", "use_accelerator": false, "weight_data_types": "int8", "weight_transformations": "TF -> OpenVINO (please refer to closed/Intel/calibration/OpenVINO)" }, { "50.00 percentile latency (ns)": 386820189726, "90.00 percentile latency (ns)": 695760617988, "95.00 percentile latency (ns)": 734460368999, "97.00 percentile latency (ns)": 749941342595, "99.00 percentile latency (ns)": 765410911946, "99.90 percentile latency (ns)": 772372750259, "Max latency (ns)": 773134010580, "Mean latency (ns)": 168457883100, "Min duration satisfied": "Yes", "Min latency (ns)": 900217258, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 109270, "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": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.power": 1431.0435963777502, "characteristics.power.normalized_per_core": 357.76089909443755, "characteristics.power.normalized_per_processor": 357.76089909443755, "characteristics.samples_per_second": 109270, "characteristics.samples_per_second.normalized_per_core": 27317.5, "characteristics.samples_per_second.normalized_per_processor": 27317.5, "characteristics.total": 50000, "ck_system": "R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "256 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 32, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8368 CPU @ 2.40GHz", "host_processors_per_node": 2, "host_storage_capacity": "543 GB", "host_storage_type": "SSD", "hw_notes": "Result Measured for Power; GPU Power Limited to 175W", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_management": "", "power_settings": "closed/Dell/power/R750xa_A100-PCIE-40GBx4_TRT_MaxQ.md", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "1x2400W", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 84480000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-40GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 128000, "task": "image classification", "task2": "image classification", "total_cores": 64, "uid": "ca805bfc34713a69", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 351140562607, "90.00 percentile latency (ns)": 632725185296, "95.00 percentile latency (ns)": 667946481866, "97.00 percentile latency (ns)": 681988540390, "99.00 percentile latency (ns)": 696087076143, "99.90 percentile latency (ns)": 702392536927, "Max latency (ns)": 703109726927, "Mean latency (ns)": 351222391034, "Min duration satisfied": "Yes", "Min latency (ns)": 261082343, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 28160.6, "Scenario": "offline", "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.accuracy": 76.03, "characteristics.good": 38015, "characteristics.samples_per_second": 28160.6, "characteristics.samples_per_second.normalized_per_core": 14080.3, "characteristics.samples_per_second.normalized_per_processor": 14080.3, "characteristics.total": 50000, "ck_system": "XE2420_A10x2_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "resnet50-v1.5", "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_cooling_type": "air", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6252 CPU @ 2.10GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "5.00.00.00", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE2420_A10x2_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_settings": "closed/Dell/power/XE2420_A10x2_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 19800000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_A10x2_TRT", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "image classification", "task2": "image classification", "total_cores": 48, "uid": "c0e7ad96c43068a5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 324810563694, "90.00 percentile latency (ns)": 584972674338, "95.00 percentile latency (ns)": 617586763333, "97.00 percentile latency (ns)": 630647317832, "99.00 percentile latency (ns)": 643699990319, "99.90 percentile latency (ns)": 649594415694, "Max latency (ns)": 650230676307, "Mean latency (ns)": 139911098772, "Min duration satisfied": "Yes", "Min latency (ns)": 149631165, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 153472, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.04, "characteristics.good": 38020, "characteristics.samples_per_second": 153472, "characteristics.samples_per_second.normalized_per_core": 19184.0, "characteristics.samples_per_second.normalized_per_processor": 19184.0, "characteristics.total": 50000, "ck_system": "DSS8440_A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/DSS8440_A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "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": 2048, "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": 99792000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT_Triton", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 151200, "task": "image classification", "task2": "image classification", "total_cores": 96, "uid": "bf35bb44748a8a40", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 362946958281, "90.00 percentile latency (ns)": 653611877762, "95.00 percentile latency (ns)": 689955393593, "97.00 percentile latency (ns)": 704440822468, "99.00 percentile latency (ns)": 718990096206, "99.90 percentile latency (ns)": 725510706290, "Max latency (ns)": 726212973666, "Mean latency (ns)": 363004464414, "Min duration satisfied": "Yes", "Min latency (ns)": 217707468, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 27264.7, "Scenario": "offline", "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 76.03, "characteristics.good": 38015, "characteristics.samples_per_second": 27264.7, "characteristics.samples_per_second.normalized_per_core": 13632.35, "characteristics.samples_per_second.normalized_per_processor": 13632.35, "characteristics.total": 50000, "ck_system": "XR12_datacenter_A10x2_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "filesystem": "", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "host_memory_capacity": "512 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) Gold 6330 CPU @ 2.00GHz", "host_processors_per_node": 1, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XR12_datacenter_A10x2_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_settings": "", "print_timestamps": 0, "problem": false, "psu_details": "", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 19800000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XR12_datacenter_A10x2_TRT", "system_name": "Dell EMC PowerEdge XR12 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "image classification", "task2": "image classification", "total_cores": 28, "uid": "994f53b5b5a0f043", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 329877049802, "90.00 percentile latency (ns)": 592670156717, "95.00 percentile latency (ns)": 625520711930, "97.00 percentile latency (ns)": 638659371094, "99.00 percentile latency (ns)": 651801212508, "99.90 percentile latency (ns)": 657834585510, "Max latency (ns)": 659547868026, "Mean latency (ns)": 137677860348, "Min duration satisfied": "Yes", "Min latency (ns)": 1074439609, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Samples per second": 145520, "Scenario": "offline", "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (7x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 76.054, "characteristics.good": 38027, "characteristics.samples_per_second": 145520, "characteristics.samples_per_second.normalized_per_core": 36380.0, "characteristics.samples_per_second.normalized_per_processor": 36380.0, "characteristics.total": 50000, "ck_system": "XE8545_A100-SXM-80GB-MIG_28x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE8545_A100-SXM-80GB-MIG_28x1g.10gb_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_management": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 95977200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE8545_A100-SXM-80GB-MIG_28x1g.10gb_TRT_Triton", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 145420, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "26399a4e66746cd9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 330167290284, "90.00 percentile latency (ns)": 594176954208, "95.00 percentile latency (ns)": 627189548341, "97.00 percentile latency (ns)": 640401018443, "99.00 percentile latency (ns)": 653583132717, "99.90 percentile latency (ns)": 659528417794, "Max latency (ns)": 660181735447, "Mean latency (ns)": 156187896519, "Min duration satisfied": "Yes", "Min latency (ns)": 101194627, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 160620, "Scenario": "offline", "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "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, "characteristics.accuracy": 76.042, "characteristics.good": 38021, "characteristics.samples_per_second": 160620, "characteristics.samples_per_second.normalized_per_core": 40155.0, "characteristics.samples_per_second.normalized_per_processor": 40155.0, "characteristics.total": 50000, "ck_system": "XE8545_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE8545_A100-SXM-80GBx4_TRT_Triton", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_management": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 106038240, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE8545_A100-SXM-80GBx4_TRT_Triton", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 160664, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "f64c2449cb64892a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 344824376945, "90.00 percentile latency (ns)": 620007175939, "95.00 percentile latency (ns)": 654428254668, "97.00 percentile latency (ns)": 668172440010, "99.00 percentile latency (ns)": 681940401926, "99.90 percentile latency (ns)": 688123258751, "Max latency (ns)": 688790675577, "Mean latency (ns)": 12860500586, "Min duration satisfied": "Yes", "Min latency (ns)": 693416840, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "LWIS_Server", "Samples per second": 161361, "Scenario": "offline", "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "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": "2.2.5", "characteristics.accuracy": 76.042, "characteristics.good": 38021, "characteristics.power": 2412.509724238026, "characteristics.power.normalized_per_core": 603.1274310595065, "characteristics.power.normalized_per_processor": 603.1274310595065, "characteristics.samples_per_second": 161361, "characteristics.samples_per_second.normalized_per_core": 40340.25, "characteristics.samples_per_second.normalized_per_processor": 40340.25, "characteristics.total": 50000, "ck_system": "XE8545_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "resnet50-v1.5", "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_cooling_type": "", "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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "4.40.40.151", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE8545_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2048, "power_management": "", "power_settings": "closed/Dell/power/XE8545_A100-SXM-80GBx4_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "4x2400W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 111144000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_v1.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE8545_A100-SXM-80GBx4_TRT", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 168400, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "2cea9641568dcd67", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 795595604521, "90.00 percentile latency (ns)": 795595604521, "95.00 percentile latency (ns)": 795595604521, "97.00 percentile latency (ns)": 795595604521, "99.00 percentile latency (ns)": 795595604521, "99.90 percentile latency (ns)": 795595604521, "Max latency (ns)": 795595604521, "Mean latency (ns)": 795595604521, "Min duration satisfied": "Yes", "Min latency (ns)": 795595604521, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "SUT\u0000", "Samples per second": 2903.49, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 76.758, "characteristics.good": 786, "characteristics.samples_per_second": 2903.49, "characteristics.samples_per_second.normalized_per_core": 45.36703125, "characteristics.samples_per_second.normalized_per_processor": 1451.745, "characteristics.total": 1024, "ck_system": "R750_Intel8352M_2S_32C", "ck_used": false, "cooling": "", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "OpenVINO", "host_memory_capacity": "512GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 32, "host_processor_frequency": "2.30GHz", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8352M CPU @ 2.30GHz", "host_processors_per_node": 2, "host_storage_capacity": "2.9TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 64, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750_Intel8352M_2S_32C", "number_of_nodes": 1, "operating_system": " Ubuntu 20.04.2 LTS (GNU/Linux 5.4.0-80-generic x86_64)", "other_software_stack": "MKL-DNN; URL: github.com/intel/mkl-dnn", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "no", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 2310000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "resnet50_int8.xml", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750_Intel8352M_2S_32C", "system_name": "Dell EMC PowerEdge R750 ICX-8352M(2S 32C)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 3500, "task": "image classification", "task2": "image classification", "total_cores": 64, "uid": "c8d44934920121b4", "use_accelerator": false, "weight_data_types": "int8", "weight_transformations": "TF -> OpenVINO (please refer to closed/Intel/calibration/OpenVINO)" } ]