[ { "50.00 percentile latency (ns)": 422426071208, "90.00 percentile latency (ns)": 622770649854, "95.00 percentile latency (ns)": 640253438821, "97.00 percentile latency (ns)": 646063078457, "99.00 percentile latency (ns)": 650840075132, "99.90 percentile latency (ns)": 652525339013, "Max latency (ns)": 652671626976, "Mean latency (ns)": 397026515413, "Min duration satisfied": "Yes", "Min latency (ns)": 1874928730, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 6735.79, "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": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 6735.79, "characteristics.samples_per_second.normalized_per_core": 1683.9475, "characteristics.samples_per_second.normalized_per_processor": 1683.9475, "ck_system": "A30x4_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 8.0.1.6, CUDA 11.3", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7302", "host_processors_per_node": 2, "host_storage_capacity": "1 TB", "host_storage_type": " 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/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Fujitsu/results/A30x4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.2 LTS", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1.6, CUDA 11.3, cuDNN 8.2.1.32, Driver 470.57, 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": 4396260, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x4_TRT", "system_name": "PRIMERGY GX2460M1 (4x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 6661, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "104e6caf365b1422", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 396871472795, "90.00 percentile latency (ns)": 586747843655, "95.00 percentile latency (ns)": 603610483669, "97.00 percentile latency (ns)": 609198815148, "99.00 percentile latency (ns)": 613692126157, "99.90 percentile latency (ns)": 615269823368, "Max latency (ns)": 615370255292, "Mean latency (ns)": 372838636816, "Min duration satisfied": "Yes", "Min latency (ns)": 1755023609, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 4290.1, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 4290.1, "characteristics.samples_per_second.normalized_per_core": 1072.525, "characteristics.samples_per_second.normalized_per_processor": 1072.525, "ck_system": "A10x4_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": "1.92 TB", "host_storage_type": "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/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/results/A10x4_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 455.45.01, 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": 2640000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "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/A10x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 4000, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "40fbb7c6d49a025a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 440447108618, "90.00 percentile latency (ns)": 650998208572, "95.00 percentile latency (ns)": 669712331393, "97.00 percentile latency (ns)": 675892919566, "99.00 percentile latency (ns)": 680808789747, "99.90 percentile latency (ns)": 682501254174, "Max latency (ns)": 682618513237, "Mean latency (ns)": 413681368648, "Min duration satisfied": "Yes", "Min latency (ns)": 1425141358, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 2154.18, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 2154.18, "characteristics.samples_per_second.normalized_per_core": 538.545, "characteristics.samples_per_second.normalized_per_processor": 538.545, "ck_system": "A10x4_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": "1.92 TB", "host_storage_type": "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/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/results/A10x4_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 455.45.01, 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": 1470480, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "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/A10x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 2228, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "e680f77285e80240", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 500088901134, "90.00 percentile latency (ns)": 738079726617, "95.00 percentile latency (ns)": 759018302075, "97.00 percentile latency (ns)": 765985480509, "99.00 percentile latency (ns)": 771612383831, "99.90 percentile latency (ns)": 773585013702, "Max latency (ns)": 773769449767, "Mean latency (ns)": 470327867254, "Min duration satisfied": "Yes", "Min latency (ns)": 3133230246, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 11429.8, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 11429.8, "characteristics.samples_per_second.normalized_per_core": 2857.45, "characteristics.samples_per_second.normalized_per_processor": 2857.45, "ck_system": "A100-PCIex4_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": "512 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 7302", "host_processors_per_node": 2, "host_storage_capacity": "1.92 TB", "host_storage_type": "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/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/results/A100-PCIex4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 455.45.01, 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": 8844000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex4_TRT", "system_name": "Fujitsu PRIMERGY GX2460 m1 (4x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 13400, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "a6c95121040061a2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 552739403413, "90.00 percentile latency (ns)": 817376330525, "95.00 percentile latency (ns)": 841069758848, "97.00 percentile latency (ns)": 848775471287, "99.00 percentile latency (ns)": 854899659852, "99.90 percentile latency (ns)": 856961039274, "Max latency (ns)": 857140302066, "Mean latency (ns)": 519339727910, "Min duration satisfied": "Yes", "Min latency (ns)": 2093986575, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 5390.02, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 5390.02, "characteristics.samples_per_second.normalized_per_core": 1347.505, "characteristics.samples_per_second.normalized_per_processor": 1347.505, "ck_system": "A100-PCIex4_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 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": 16, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7302", "host_processors_per_node": 2, "host_storage_capacity": "1.92 TB", "host_storage_type": "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/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/results/A100-PCIex4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 455.45.01, 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": 4620000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex4_TRT", "system_name": "Fujitsu PRIMERGY GX2460 m1 (4x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 7000, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "d52b144ecb9d8215", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 469974972896, "90.00 percentile latency (ns)": 693610048144, "95.00 percentile latency (ns)": 713085237710, "97.00 percentile latency (ns)": 719562285133, "99.00 percentile latency (ns)": 724829506630, "99.90 percentile latency (ns)": 726710655935, "Max latency (ns)": 726884882928, "Mean latency (ns)": 441803876897, "Min duration satisfied": "Yes", "Min latency (ns)": 1970903662, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 5811.1, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 5811.1, "characteristics.samples_per_second.normalized_per_core": 2905.55, "characteristics.samples_per_second.normalized_per_processor": 2905.55, "ck_system": "A100-PCIex2_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": "384 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": "1.92 TB", "host_storage_type": "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": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/results/A100-PCIex2_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 455.45.01, 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": 4224000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "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/A100-PCIex2_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 6400, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "e458308204764568", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 478331537209, "90.00 percentile latency (ns)": 704367757675, "95.00 percentile latency (ns)": 724646118281, "97.00 percentile latency (ns)": 731141721170, "99.00 percentile latency (ns)": 736367192829, "99.90 percentile latency (ns)": 738134578301, "Max latency (ns)": 738271416407, "Mean latency (ns)": 449104014071, "Min duration satisfied": "Yes", "Min latency (ns)": 2268404639, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 2807.1, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.samples_per_second": 2807.1, "characteristics.samples_per_second.normalized_per_core": 1403.55, "characteristics.samples_per_second.normalized_per_processor": 1403.55, "ck_system": "A100-PCIex2_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": "384 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": "1.92 TB", "host_storage_type": "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": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Fujitsu/results/A100-PCIex2_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 455.45.01, 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": 2072400, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "preview", "submitter": "Fujitsu", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Fujitsu", "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/A100-PCIex2_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 3140, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "f34b659fa4bc7eec", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" } ]