[ { "50.00 percentile latency (ns)": 428526945756, "90.00 percentile latency (ns)": 631391479946, "95.00 percentile latency (ns)": 649062346690, "97.00 percentile latency (ns)": 654964016673, "99.00 percentile latency (ns)": 659844890839, "99.90 percentile latency (ns)": 661562700545, "Max latency (ns)": 661712393544, "Mean latency (ns)": 403681762034, "Min duration satisfied": "Yes", "Min latency (ns)": 6458584228, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 30221.6, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "", "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, "boot_firmware_version": "", "characteristics.samples_per_second": 30221.6, "characteristics.samples_per_second.normalized_per_core": 3777.7, "characteristics.samples_per_second.normalized_per_processor": 3777.7, "ck_system": "NF5488A5_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "Liquid", "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": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "11.7 TB + 448GB*2", "host_storage_type": "NVMe SSD + M.2", "hw_notes": "NVIDIA A100-SXM-80GB-500W", "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NF5488A5_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, 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": 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": 19998000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30300, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "a1e13c8437d9dca6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 427323102073, "90.00 percentile latency (ns)": 628053911665, "95.00 percentile latency (ns)": 645769747703, "97.00 percentile latency (ns)": 651494758065, "99.00 percentile latency (ns)": 656145768665, "99.90 percentile latency (ns)": 657656848146, "Max latency (ns)": 657801841809, "Mean latency (ns)": 401645385989, "Min duration satisfied": "Yes", "Min latency (ns)": 4099985319, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 15050.1, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "", "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, "boot_firmware_version": "", "characteristics.samples_per_second": 15050.1, "characteristics.samples_per_second.normalized_per_core": 1881.2625, "characteristics.samples_per_second.normalized_per_processor": 1881.2625, "ck_system": "NF5488A5_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "Liquid", "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": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "11.7 TB + 448GB*2", "host_storage_type": "NVMe SSD + M.2", "hw_notes": "NVIDIA A100-SXM-80GB-500W", "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NF5488A5_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, 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": 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": 9900000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "ffcbfb4135bad927", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 451499959447, "90.00 percentile latency (ns)": 662903146484, "95.00 percentile latency (ns)": 681237979738, "97.00 percentile latency (ns)": 687287004096, "99.00 percentile latency (ns)": 692239925494, "99.90 percentile latency (ns)": 694016880478, "Max latency (ns)": 694157800896, "Mean latency (ns)": 424775863234, "Min duration satisfied": "Yes", "Min latency (ns)": 7090132536, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 28523.8, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4-40GB", "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": 28523.8, "characteristics.samples_per_second.normalized_per_core": 3565.475, "characteristics.samples_per_second.normalized_per_processor": 3565.475, "ck_system": "NF5488M6_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, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.2", "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", "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488M6_A100-SXM4-40GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0; 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": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "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/NF5488M6_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": "7ab9c626abe8b571", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 467323348787, "90.00 percentile latency (ns)": 685883813207, "95.00 percentile latency (ns)": 705094942457, "97.00 percentile latency (ns)": 711334295480, "99.00 percentile latency (ns)": 716385110409, "99.90 percentile latency (ns)": 718051361273, "Max latency (ns)": 718175401237, "Mean latency (ns)": 438929434007, "Min duration satisfied": "Yes", "Min latency (ns)": 5010152241, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 13784.9, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4-40GB", "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": 13784.9, "characteristics.samples_per_second.normalized_per_core": 1723.1125, "characteristics.samples_per_second.normalized_per_processor": 1723.1125, "ck_system": "NF5488M6_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, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.2", "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", "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488M6_A100-SXM4-40GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0; 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": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "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/NF5488M6_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": "d74f569f0c85b135", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 427170702617, "90.00 percentile latency (ns)": 629310347948, "95.00 percentile latency (ns)": 646942893345, "97.00 percentile latency (ns)": 652833288830, "99.00 percentile latency (ns)": 657677734169, "99.90 percentile latency (ns)": 659400442600, "Max latency (ns)": 659560330918, "Mean latency (ns)": 402281191073, "Min duration satisfied": "Yes", "Min latency (ns)": 6295791316, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 30020, "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": 30020, "characteristics.samples_per_second.normalized_per_core": 3752.5, "characteristics.samples_per_second.normalized_per_processor": 3752.5, "ck_system": "NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "Liquid", "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, CUDA 11.2", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.2", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09", "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": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5 (Liquid cooling)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "32424ab7b28bc19e", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 433686644665, "90.00 percentile latency (ns)": 636992631948, "95.00 percentile latency (ns)": 654978626120, "97.00 percentile latency (ns)": 660802520625, "99.00 percentile latency (ns)": 665490760790, "99.90 percentile latency (ns)": 667109049234, "Max latency (ns)": 667274023916, "Mean latency (ns)": 407490207996, "Min duration satisfied": "Yes", "Min latency (ns)": 3648751938, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 14836.5, "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": 14836.5, "characteristics.samples_per_second.normalized_per_core": 1854.5625, "characteristics.samples_per_second.normalized_per_processor": 1854.5625, "ck_system": "NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "Liquid", "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, CUDA 11.2", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.2", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09", "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": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5 (Liquid cooling)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "cb79de09134489b7", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 520369073842, "90.00 percentile latency (ns)": 766306122463, "95.00 percentile latency (ns)": 787742523768, "97.00 percentile latency (ns)": 794911587385, "99.00 percentile latency (ns)": 800706382763, "99.90 percentile latency (ns)": 802761573663, "Max latency (ns)": 802941010092, "Mean latency (ns)": 489541089822, "Min duration satisfied": "Yes", "Min latency (ns)": 6185836528, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 23015.4, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 23015.4, "characteristics.samples_per_second.normalized_per_core": 2876.925, "characteristics.samples_per_second.normalized_per_processor": 2876.925, "ck_system": "NF5468M6_A100-PCIex8_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, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5468M6_A100-PCIex8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; 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": 18480000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "Preview", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "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/NF5468M6_A100-PCIex8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 28000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "6deb6eb104741099", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 454136316646, "90.00 percentile latency (ns)": 668651540960, "95.00 percentile latency (ns)": 687442197002, "97.00 percentile latency (ns)": 693724608447, "99.00 percentile latency (ns)": 698860375712, "99.90 percentile latency (ns)": 700672834568, "Max latency (ns)": 700844163480, "Mean latency (ns)": 427465040852, "Min duration satisfied": "Yes", "Min latency (ns)": 6185711506, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 28251.6, "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": 28251.6, "characteristics.samples_per_second.normalized_per_core": 3531.45, "characteristics.samples_per_second.normalized_per_processor": 3531.45, "ck_system": "NF5488A5_A100-SXM-80GBx8_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, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.2", "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": "12 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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488A5_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.2", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09", "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": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 30000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "b89e42b794502ab9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 463284879701, "90.00 percentile latency (ns)": 681571220164, "95.00 percentile latency (ns)": 700872215223, "97.00 percentile latency (ns)": 707155436038, "99.00 percentile latency (ns)": 712206384137, "99.90 percentile latency (ns)": 713933465284, "Max latency (ns)": 714072056690, "Mean latency (ns)": 435324370929, "Min duration satisfied": "Yes", "Min latency (ns)": 3351101176, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 13864.1, "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": 13864.1, "characteristics.samples_per_second.normalized_per_core": 1733.0125, "characteristics.samples_per_second.normalized_per_processor": 1733.0125, "ck_system": "NF5488A5_A100-SXM-80GBx8_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, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.2", "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": "12 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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488A5_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.2", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09", "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": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "0ba15d9db6c51a1d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 43609797127, "90.00 percentile latency (ns)": 64233456625, "95.00 percentile latency (ns)": 66053758944, "97.00 percentile latency (ns)": 66615540572, "99.00 percentile latency (ns)": 67102737137, "99.90 percentile latency (ns)": 67275650418, "Max latency (ns)": 67299438632, "Mean latency (ns)": 41118712285, "Min duration satisfied": "Yes", "Min latency (ns)": 1218972437, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 26674.8, "Scenario": "offline", "accelerator_frequency": "1410MHz", "accelerator_host_interconnect": "PCIe Gen4 64 GB/s", "accelerator_interconnect": "NVLink 600 GB/s", "accelerator_interconnect_topology": "Mesh", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "L1: 108x192KB, L2: 40MB per chip", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 26674.8, "characteristics.samples_per_second.normalized_per_core": 3334.35, "characteristics.samples_per_second.normalized_per_processor": 3334.35, "ck_system": "DGX-A100_A100-SXM4x8_TRT", "ck_used": false, "cooling": "Air-cooled", "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, CUDA 11.0.221", "host_memory_capacity": "1 TB", "host_memory_configuration": "16 x 64GB DDR4 2933 MT/s", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "L1: 32KB I + 32KB D per core, L2: 512KB per core, L3: 256MB per chip", "host_processor_core_count": 64, "host_processor_frequency": "2.25GHz", "host_processor_interconnect": "4x 16GT/s xGMI", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "11.7TB + 445.7GB", "host_storage_type": "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)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Inspur/results/DGX-A100_A100-SXM4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0.221, cuDNN 8.0.2", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 10833, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1795200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "on-premise", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT", "system_name": "NF5488A5", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 27200, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "69313c50484742ab", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 46781655818, "90.00 percentile latency (ns)": 68512497897, "95.00 percentile latency (ns)": 70457435576, "97.00 percentile latency (ns)": 71191753320, "99.00 percentile latency (ns)": 71618238767, "99.90 percentile latency (ns)": 71786724292, "Max latency (ns)": 71787206396, "Mean latency (ns)": 43996803788, "Min duration satisfied": "Yes", "Min latency (ns)": 1577640045, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 13790.8, "Scenario": "offline", "accelerator_frequency": "1410MHz", "accelerator_host_interconnect": "PCIe Gen4 64 GB/s", "accelerator_interconnect": "NVLink 600 GB/s", "accelerator_interconnect_topology": "Mesh", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "L1: 108x192KB, L2: 40MB per chip", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 13790.8, "characteristics.samples_per_second.normalized_per_core": 1723.85, "characteristics.samples_per_second.normalized_per_processor": 1723.85, "ck_system": "DGX-A100_A100-SXM4x8_TRT", "ck_used": false, "cooling": "Air-cooled", "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, CUDA 11.0.221", "host_memory_capacity": "1 TB", "host_memory_configuration": "16 x 64GB DDR4 2933 MT/s", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "L1: 32KB I + 32KB D per core, L2: 512KB per core, L3: 256MB per chip", "host_processor_core_count": 64, "host_processor_frequency": "2.25GHz", "host_processor_interconnect": "4x 16GT/s xGMI", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "11.7TB + 445.7GB", "host_storage_type": "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)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Inspur/results/DGX-A100_A100-SXM4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0.221, cuDNN 8.0.2", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 10833, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 990000, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "on-premise", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT", "system_name": "NF5488A5", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 15000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "8d388c45dfb95b78", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" } ]