[ { "50.00 percentile latency (ns)": 467708114793, "90.00 percentile latency (ns)": 687613620423, "95.00 percentile latency (ns)": 706674873244, "97.00 percentile latency (ns)": 713082634304, "99.00 percentile latency (ns)": 718179652372, "99.90 percentile latency (ns)": 720040580192, "Max latency (ns)": 720174618449, "Mean latency (ns)": 439698755404, "Min duration satisfied": "Yes", "Min latency (ns)": 5026458130, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 12463.6, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 12463.6, "characteristics.samples_per_second.normalized_per_core": 3115.9, "characteristics.samples_per_second.normalized_per_processor": 3115.9, "ck_system": "SYS-120GQ-TNRT_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.4", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 18, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6354", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/SYS-120GQ-TNRT_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.1", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, 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": 10833, "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": 8976000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/SYS-120GQ-TNRT_TRT", "system_name": "Supermicro SYS-120GQ-TNRT (4x A100-PCIe-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 13600, "task": "NLP", "task2": "nlp", "total_cores": 36, "uid": "5b4a337450ca6000", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 844145786300, "90.00 percentile latency (ns)": 1241031580783, "95.00 percentile latency (ns)": 1275848001758, "97.00 percentile latency (ns)": 1287152602192, "99.00 percentile latency (ns)": 1296202575471, "99.90 percentile latency (ns)": 1299222050598, "Max latency (ns)": 1299510040578, "Mean latency (ns)": 792729516548, "Min duration satisfied": "Yes", "Min latency (ns)": 6235481362, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 6246.97, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 6246.97, "characteristics.samples_per_second.normalized_per_core": 1561.7425, "characteristics.samples_per_second.normalized_per_processor": 1561.7425, "ck_system": "SYS-120GQ-TNRT_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.4", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 18, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6354", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/SYS-120GQ-TNRT_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.1", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, 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": 10833, "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": 8118000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/SYS-120GQ-TNRT_TRT", "system_name": "Supermicro SYS-120GQ-TNRT (4x A100-PCIe-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 12300, "task": "NLP", "task2": "nlp", "total_cores": 36, "uid": "afe837487e8b7033", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 454547750213, "90.00 percentile latency (ns)": 681201845838, "95.00 percentile latency (ns)": 701007921144, "97.00 percentile latency (ns)": 707560367691, "99.00 percentile latency (ns)": 712801686284, "99.90 percentile latency (ns)": 714649513675, "Max latency (ns)": 714762831446, "Mean latency (ns)": 429257809545, "Min duration satisfied": "Yes", "Min latency (ns)": 5365313967, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 12004, "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.samples_per_second": 12004, "characteristics.samples_per_second.normalized_per_core": 1500.5, "characteristics.samples_per_second.normalized_per_processor": 1500.5, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": "7 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 8580000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "SUPERMICRO", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Supermicro AS-4124GS-TNR (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 13000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "14d4e218f6e2e07d", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 576716701463, "90.00 percentile latency (ns)": 861053456625, "95.00 percentile latency (ns)": 886311651866, "97.00 percentile latency (ns)": 894623472487, "99.00 percentile latency (ns)": 900933785343, "99.90 percentile latency (ns)": 903142667169, "Max latency (ns)": 903383040008, "Mean latency (ns)": 543394131957, "Min duration satisfied": "Yes", "Min latency (ns)": 6937978796, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 5932.37, "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.samples_per_second": 5932.37, "characteristics.samples_per_second.normalized_per_core": 741.54625, "characteristics.samples_per_second.normalized_per_processor": 741.54625, "ck_system": "A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": "7 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 5359199, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "SUPERMICRO", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Supermicro AS-4124GS-TNR (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 8120, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "8a5788087e497ed3", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 442945158478, "90.00 percentile latency (ns)": 661780225649, "95.00 percentile latency (ns)": 681000283612, "97.00 percentile latency (ns)": 687375507718, "99.00 percentile latency (ns)": 692626287613, "99.90 percentile latency (ns)": 694489729578, "Max latency (ns)": 694654008504, "Mean latency (ns)": 417818861024, "Min duration satisfied": "Yes", "Min latency (ns)": 3403207267, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 12351.5, "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.samples_per_second": 12351.5, "characteristics.samples_per_second.normalized_per_core": 1543.9375, "characteristics.samples_per_second.normalized_per_processor": 1543.9375, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": "7 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 8580000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "SUPERMICRO", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT_Triton", "system_name": "Supermicro AS-4124GS-TNR (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 13000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "410348f3ba480199", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 568056839524, "90.00 percentile latency (ns)": 846080116644, "95.00 percentile latency (ns)": 870466956384, "97.00 percentile latency (ns)": 878523248480, "99.00 percentile latency (ns)": 884896149333, "99.90 percentile latency (ns)": 887076118173, "Max latency (ns)": 887167519795, "Mean latency (ns)": 534676807513, "Min duration satisfied": "Yes", "Min latency (ns)": 4005184341, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "Triton_Server", "Samples per second": 6040.8, "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.samples_per_second": 6040.8, "characteristics.samples_per_second.normalized_per_core": 755.1, "characteristics.samples_per_second.normalized_per_processor": 755.1, "ck_system": "A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": "7 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0, Triton 21.07", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 5359199, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "SUPERMICRO", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT_Triton", "system_name": "Supermicro AS-4124GS-TNR (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 8120, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "8f4e4c249c164141", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 521277400915, "90.00 percentile latency (ns)": 764993840457, "95.00 percentile latency (ns)": 786281412224, "97.00 percentile latency (ns)": 793198994699, "99.00 percentile latency (ns)": 798945036305, "99.90 percentile latency (ns)": 800961755566, "Max latency (ns)": 801141072286, "Mean latency (ns)": 490156636458, "Min duration satisfied": "Yes", "Min latency (ns)": 8519734474, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 28010, "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": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 28010, "characteristics.samples_per_second.normalized_per_core": 2801.0, "characteristics.samples_per_second.normalized_per_processor": 2801.0, "ck_system": "SYS-420GP-TNR_TRT-1", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "CUDA 11.4", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 36, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/SYS-420GP-TNR_TRT-1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "CUDA 11.4, Driver 470.57.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 22440000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/SYS-420GP-TNR_TRT-1", "system_name": "Supermicro SYS-420GP-TNR [10xA100-PCIe-40GB]", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 34000, "task": "NLP", "task2": "nlp", "total_cores": 72, "uid": "d879e8fedfd69035", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 421279291420, "90.00 percentile latency (ns)": 619444061702, "95.00 percentile latency (ns)": 636927643379, "97.00 percentile latency (ns)": 642590069774, "99.00 percentile latency (ns)": 647103677051, "99.90 percentile latency (ns)": 648575533245, "Max latency (ns)": 648637557641, "Mean latency (ns)": 395838144901, "Min duration satisfied": "Yes", "Min latency (ns)": 5839771255, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 14204.5, "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": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 14204.5, "characteristics.samples_per_second.normalized_per_core": 1420.45, "characteristics.samples_per_second.normalized_per_processor": 1420.45, "ck_system": "SYS-420GP-TNR_TRT-1", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "CUDA 11.4", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 36, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Supermicro/results/SYS-420GP-TNR_TRT-1", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "CUDA 11.4, Driver 470.57.02", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 9213600, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/SYS-420GP-TNR_TRT-1", "system_name": "Supermicro SYS-420GP-TNR [10xA100-PCIe-40GB]", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 13960, "task": "NLP", "task2": "nlp", "total_cores": 72, "uid": "5f9b9e8a47718fb5", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 455966526291, "90.00 percentile latency (ns)": 669911665308, "95.00 percentile latency (ns)": 688452423649, "97.00 percentile latency (ns)": 694585367164, "99.00 percentile latency (ns)": 699670575188, "99.90 percentile latency (ns)": 701367258477, "Max latency (ns)": 701586906872, "Mean latency (ns)": 428982723859, "Min duration satisfied": "Yes", "Min latency (ns)": 6666967119, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 28221.7, "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-SXM-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 28221.7, "characteristics.samples_per_second.normalized_per_core": 3527.7125, "characteristics.samples_per_second.normalized_per_processor": 3527.7125, "ck_system": "AS-4124GO-NART_A100-SXM4-40GBx8_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-4124GO-NART_A100-SXM4-40GBx8_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 460.32.03, 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": 10833, "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": 19800000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-4124GO-NART_A100-SXM4-40GBx8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "47e0f7491482f17c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 472681787492, "90.00 percentile latency (ns)": 693579593094, "95.00 percentile latency (ns)": 713012295421, "97.00 percentile latency (ns)": 719321541659, "99.00 percentile latency (ns)": 724416261029, "99.90 percentile latency (ns)": 726091694887, "Max latency (ns)": 726247531942, "Mean latency (ns)": 443988112567, "Min duration satisfied": "Yes", "Min latency (ns)": 5310330822, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 13631.7, "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-SXM-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 13631.7, "characteristics.samples_per_second.normalized_per_core": 1703.9625, "characteristics.samples_per_second.normalized_per_processor": 1703.9625, "ck_system": "AS-4124GO-NART_A100-SXM4-40GBx8_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-4124GO-NART_A100-SXM4-40GBx8_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 460.32.03, 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": 10833, "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": 9900000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-4124GO-NART_A100-SXM4-40GBx8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "682120a0c10f469a", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 454355874805, "90.00 percentile latency (ns)": 668452075041, "95.00 percentile latency (ns)": 687069428007, "97.00 percentile latency (ns)": 693244816340, "99.00 percentile latency (ns)": 698250479679, "99.90 percentile latency (ns)": 700029946320, "Max latency (ns)": 700171498717, "Mean latency (ns)": 427185728361, "Min duration satisfied": "Yes", "Min latency (ns)": 4002506972, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 14139.4, "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-SXM-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.samples_per_second": 14139.4, "characteristics.samples_per_second.normalized_per_core": 3534.85, "characteristics.samples_per_second.normalized_per_processor": 3534.85, "ck_system": "AS-2124GQ-NART_A100-SXM4-40GBx4_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-2124GQ-NART_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 460.32.03, 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": 10833, "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": 9900000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-2124GQ-NART_A100-SXM4-40GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "cf1644d4bd697dfd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 472161978113, "90.00 percentile latency (ns)": 693368209361, "95.00 percentile latency (ns)": 713026935158, "97.00 percentile latency (ns)": 719226999941, "99.00 percentile latency (ns)": 724261730072, "99.90 percentile latency (ns)": 726003231228, "Max latency (ns)": 726074905377, "Mean latency (ns)": 443421005977, "Min duration satisfied": "Yes", "Min latency (ns)": 4116386089, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 6817.48, "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-SXM-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.samples_per_second": 6817.48, "characteristics.samples_per_second.normalized_per_core": 1704.37, "characteristics.samples_per_second.normalized_per_processor": 1704.37, "ck_system": "AS-2124GQ-NART_A100-SXM4-40GBx4_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-2124GQ-NART_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 460.32.03, 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": 10833, "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": 4950000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-2124GQ-NART_A100-SXM4-40GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7500, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "19d1cfb39ec737cc", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 393864865586, "90.00 percentile latency (ns)": 580570040775, "95.00 percentile latency (ns)": 596650032042, "97.00 percentile latency (ns)": 602121005024, "99.00 percentile latency (ns)": 606437089793, "99.90 percentile latency (ns)": 607995226416, "Max latency (ns)": 608115496907, "Mean latency (ns)": 370548040915, "Min duration satisfied": "Yes", "Min latency (ns)": 3142855984, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 14760.4, "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.samples_per_second": 14760.4, "characteristics.samples_per_second.normalized_per_core": 3690.1, "characteristics.samples_per_second.normalized_per_processor": 3690.1, "ck_system": "AS-2124GQ-NART_A100-SXM-80GBx4_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-2124GQ-NART_A100-SXM-80GBx4_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 460.32.03, 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": 10833, "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": 8976000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-2124GQ-NART_A100-SXM-80GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 13600, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "dfe94717ee415ea1", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 424798341695, "90.00 percentile latency (ns)": 623775138100, "95.00 percentile latency (ns)": 641288678324, "97.00 percentile latency (ns)": 646925551690, "99.00 percentile latency (ns)": 651368726881, "99.90 percentile latency (ns)": 652879097319, "Max latency (ns)": 652991078622, "Mean latency (ns)": 398936139186, "Min duration satisfied": "Yes", "Min latency (ns)": 3492552759, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 7075.13, "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.samples_per_second": 7075.13, "characteristics.samples_per_second.normalized_per_core": 1768.7825, "characteristics.samples_per_second.normalized_per_processor": 1768.7825, "ck_system": "AS-2124GQ-NART_A100-SXM-80GBx4_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "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/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-2124GQ-NART_A100-SXM-80GBx4_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 460.32.03, 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": 10833, "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": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-2124GQ-NART_A100-SXM-80GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "8f414c80a90ca694", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 454948419534, "90.00 percentile latency (ns)": 669891850422, "95.00 percentile latency (ns)": 688634151057, "97.00 percentile latency (ns)": 694940812054, "99.00 percentile latency (ns)": 700046273319, "99.90 percentile latency (ns)": 701852223129, "Max latency (ns)": 702007225037, "Mean latency (ns)": 428055491764, "Min duration satisfied": "Yes", "Min latency (ns)": 5185271179, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 28204.8, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 28204.8, "characteristics.samples_per_second.normalized_per_core": 3525.6, "characteristics.samples_per_second.normalized_per_processor": 3525.6, "ck_system": "AS-4214GO-NART_A100-SXM-80GBx8_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-4214GO-NART_A100-SXM-80GBx8_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 7.2.3, 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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 19800000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-4214GO-NART_A100-SXM-80GBx8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "f5659c8d41b08769", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 467073635942, "90.00 percentile latency (ns)": 686794459963, "95.00 percentile latency (ns)": 706226490653, "97.00 percentile latency (ns)": 712563909169, "99.00 percentile latency (ns)": 717651210718, "99.90 percentile latency (ns)": 719377218163, "Max latency (ns)": 719534782917, "Mean latency (ns)": 438753085456, "Min duration satisfied": "Yes", "Min latency (ns)": 2804393085, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 13758.9, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 13758.9, "characteristics.samples_per_second.normalized_per_core": 1719.8625, "characteristics.samples_per_second.normalized_per_processor": 1719.8625, "ck_system": "AS-4214GO-NART_A100-SXM-80GBx8_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/AS-4214GO-NART_A100-SXM-80GBx8_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 7.2.3, 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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 9900000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/AS-4214GO-NART_A100-SXM-80GBx8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "7aaa7270e447adbd", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 511539030080, "90.00 percentile latency (ns)": 751498798876, "95.00 percentile latency (ns)": 772270709643, "97.00 percentile latency (ns)": 779220194666, "99.00 percentile latency (ns)": 784744138802, "99.90 percentile latency (ns)": 786743408407, "Max latency (ns)": 786915929242, "Mean latency (ns)": 480880468954, "Min duration satisfied": "Yes", "Min latency (ns)": 5583742590, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 22813.1, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 22813.1, "characteristics.samples_per_second.normalized_per_core": 2851.6375, "characteristics.samples_per_second.normalized_per_processor": 2851.6375, "ck_system": "A100-PCIex8_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": 16, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7313", "host_processors_per_node": 2, "host_storage_capacity": "7 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": 17952000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Supermicro AS-4124GS-TNR (8x A100-PCIe)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 27200, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "9f4fcb805c707c63", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 455306284521, "90.00 percentile latency (ns)": 670220839121, "95.00 percentile latency (ns)": 689023426239, "97.00 percentile latency (ns)": 695265254359, "99.00 percentile latency (ns)": 700382165496, "99.90 percentile latency (ns)": 702217966026, "Max latency (ns)": 702351606260, "Mean latency (ns)": 428457206916, "Min duration satisfied": "Yes", "Min latency (ns)": 5787916686, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 28191, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 28191, "characteristics.samples_per_second.normalized_per_core": 3523.875, "characteristics.samples_per_second.normalized_per_processor": 3523.875, "ck_system": "SYS-420GP-TNAR_A100-SXM-80GBx8_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "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": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/SYS-420GP-TNAR_A100-SXM-80GBx8_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 7.2.3, 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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 19800000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/SYS-420GP-TNAR_A100-SXM-80GBx8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "47d19c1d5bd6cda1", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 467172196558, "90.00 percentile latency (ns)": 686909917596, "95.00 percentile latency (ns)": 706336771505, "97.00 percentile latency (ns)": 712703622620, "99.00 percentile latency (ns)": 717759716758, "99.90 percentile latency (ns)": 719482426769, "Max latency (ns)": 719652491789, "Mean latency (ns)": 438893623752, "Min duration satisfied": "Yes", "Min latency (ns)": 3152306740, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 13756.6, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 13756.6, "characteristics.samples_per_second.normalized_per_core": 1719.575, "characteristics.samples_per_second.normalized_per_processor": 1719.575, "ck_system": "SYS-420GP-TNAR_A100-SXM-80GBx8_TRT", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "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": "3.5 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Supermicro/results/SYS-420GP-TNAR_A100-SXM-80GBx8_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 7.2.3, 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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 9900000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Supermicro", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Supermicro", "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/SYS-420GP-TNAR_A100-SXM-80GBx8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "2181600402e855fa", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" } ]