[ { "50.00 percentile latency (ns)": 1420862, "90.00 percentile latency (ns)": 1662836, "90th percentile latency (ns)": 1662836, "95.00 percentile latency (ns)": 1773145, "97.00 percentile latency (ns)": 1826459, "99.00 percentile latency (ns)": 1839313, "99.90 percentile latency (ns)": 1855695, "Max latency (ns)": 10327074, "Mean latency (ns)": 1478355, "Min duration satisfied": "Yes", "Min latency (ns)": 1314508, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 670.46, "QPS w/o loadgen overhead": 676.43, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "", "accelerator_model_name": "NVIDIA A100-PCIe", "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.90th_percentile_latency_ms": 1.662836, "characteristics.90th_percentile_latency_ns": 1662836.0, "characteristics.90th_percentile_latency_s": 0.001662836, "characteristics.90th_percentile_latency_us": 1662.836, "characteristics.power": 0.7370190450479738, "characteristics.power.normalized_per_core": 0.3685095225239869, "characteristics.power.normalized_per_processor": 0.3685095225239869, "ck_system": "NE5260M5_A100-PCIex2_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "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": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6258R", "host_processors_per_node": 2, "host_storage_capacity": "2.9 TB * 2", "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": 1024, "mlperf_version": 1.1, "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_v1.1/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NE5260M5_A100-PCIex2_TRT_MaxQ", "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": 1, "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/NE5260M5_A100-PCIex2_TRT_MaxQ", "system_name": "NE5260M5 (2x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 56, "uid": "e79cb7c2dafa4202", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1431312, "90.00 percentile latency (ns)": 1555270, "90th percentile latency (ns)": 1555270, "95.00 percentile latency (ns)": 1661084, "97.00 percentile latency (ns)": 1739808, "99.00 percentile latency (ns)": 1748010, "99.90 percentile latency (ns)": 1753269, "Max latency (ns)": 4656529, "Mean latency (ns)": 1459394, "Min duration satisfied": "Yes", "Min latency (ns)": 1339121, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 682.94, "QPS w/o loadgen overhead": 685.22, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "", "accelerator_model_name": "NVIDIA A100-PCIe", "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.90th_percentile_latency_ms": 1.55527, "characteristics.90th_percentile_latency_ns": 1555270.0, "characteristics.90th_percentile_latency_s": 0.00155527, "characteristics.90th_percentile_latency_us": 1555.27, "ck_system": "NE5260M5_A100-PCIex2_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "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": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6258R", "host_processors_per_node": 2, "host_storage_capacity": "2.9 TB * 2", "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": 1024, "mlperf_version": 1.1, "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_v1.1/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NE5260M5_A100-PCIex2_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": 1, "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/NE5260M5_A100-PCIex2_TRT", "system_name": "NE5260M5 (2x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 56, "uid": "279656ed49e4ed68", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1535016, "90.00 percentile latency (ns)": 1592422, "90th percentile latency (ns)": 1592422, "95.00 percentile latency (ns)": 1700414, "97.00 percentile latency (ns)": 1769643, "99.00 percentile latency (ns)": 2854056, "99.90 percentile latency (ns)": 2868373, "Max latency (ns)": 6317505, "Mean latency (ns)": 1578010, "Min duration satisfied": "Yes", "Min latency (ns)": 1464010, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 631.73, "QPS w/o loadgen overhead": 633.71, "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "", "accelerator_model_name": "NVIDIA A100-PCIe", "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.90th_percentile_latency_ms": 1.592422, "characteristics.90th_percentile_latency_ns": 1592422.0, "characteristics.90th_percentile_latency_s": 0.001592422, "characteristics.90th_percentile_latency_us": 1592.422, "ck_system": "NE5260M5_A100-PCIex2_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "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": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6258R", "host_processors_per_node": 2, "host_storage_capacity": "2.9 TB * 2", "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": 1024, "mlperf_version": 1.1, "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_v1.1/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NE5260M5_A100-PCIex2_TRT_Triton", "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": 1, "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/NE5260M5_A100-PCIex2_TRT_Triton", "system_name": "NE5260M5 (2x A100-PCIe, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 56, "uid": "475439899d189844", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1422407, "90.00 percentile latency (ns)": 1544876, "90th percentile latency (ns)": 1544876, "95.00 percentile latency (ns)": 1651637, "97.00 percentile latency (ns)": 1726772, "99.00 percentile latency (ns)": 1740331, "99.90 percentile latency (ns)": 1743410, "Max latency (ns)": 3113653, "Mean latency (ns)": 1450752, "Min duration satisfied": "Yes", "Min latency (ns)": 1327974, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 687.28, "QPS w/o loadgen overhead": 689.3, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.90th_percentile_latency_ms": 1.544876, "characteristics.90th_percentile_latency_ns": 1544876.0, "characteristics.90th_percentile_latency_s": 0.001544876, "characteristics.90th_percentile_latency_us": 1544.876, "ck_system": "NF5688M6_A100-SXM-80GBx1_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "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": 32, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8358", "host_processors_per_node": 2, "host_storage_capacity": "28 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": 1024, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "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/NF5688M6_A100-SXM-80GBx1_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 1.0.0", "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": 1, "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/NF5688M6_A100-SXM-80GBx1_TRT", "system_name": "NF5688M6 (8x A100-SXM-80GB, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 64, "uid": "e75272414599a51c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1549819, "90.00 percentile latency (ns)": 1613546, "90th percentile latency (ns)": 1613546, "95.00 percentile latency (ns)": 1731150, "97.00 percentile latency (ns)": 1801209, "99.00 percentile latency (ns)": 2838667, "99.90 percentile latency (ns)": 3966958, "Max latency (ns)": 15438602, "Mean latency (ns)": 1600400, "Min duration satisfied": "Yes", "Min latency (ns)": 1474830, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 614.71, "QPS w/o loadgen overhead": 624.84, "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "singlestream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.90th_percentile_latency_ms": 1.613546, "characteristics.90th_percentile_latency_ns": 1613546.0, "characteristics.90th_percentile_latency_s": 0.001613546, "characteristics.90th_percentile_latency_us": 1613.546, "ck_system": "NF5488A5_A100-SXM-80GBx1_TRT_Triton", "ck_used": false, "cooling": "Liquid", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "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": 1024, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "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-80GBx1_TRT_Triton", "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, 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": 1, "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-80GBx1_TRT_Triton", "system_name": "NF5488A5 (1x A100-SXM-80GB, TensorRT, Triton)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "9b24272754531ba1", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1962966, "90.00 percentile latency (ns)": 2181874, "90th percentile latency (ns)": 2181874, "95.00 percentile latency (ns)": 2263438, "97.00 percentile latency (ns)": 2341706, "99.00 percentile latency (ns)": 2350779, "99.90 percentile latency (ns)": 2360041, "Max latency (ns)": 7312975, "Mean latency (ns)": 2001967, "Min duration satisfied": "Yes", "Min latency (ns)": 1833766, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 498.27, "QPS w/o loadgen overhead": 499.51, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 2.181874, "characteristics.90th_percentile_latency_ns": 2181874.0, "characteristics.90th_percentile_latency_s": 0.002181874, "characteristics.90th_percentile_latency_us": 2181.874, "ck_system": "NE5260M5_A100-PCIex2_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA Version: 11.2", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6258R CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "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": 1024, "mlperf_version": 1.0, "normalize_cores": 2, "normalize_processors": 2, "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/NE5260M5_A100-PCIex2_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", "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": 1, "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/NE5260M5_A100-PCIex2_TRT", "system_name": "NE5260M5", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 56, "uid": "d424593b419d6a41", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1585058, "90.00 percentile latency (ns)": 1720624, "90th percentile latency (ns)": 1720624, "95.00 percentile latency (ns)": 1780200, "97.00 percentile latency (ns)": 1846981, "99.00 percentile latency (ns)": 1855237, "99.90 percentile latency (ns)": 1898722, "Max latency (ns)": 5317034, "Mean latency (ns)": 1611761, "Min duration satisfied": "Yes", "Min latency (ns)": 1496886, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 617.91, "QPS w/o loadgen overhead": 620.44, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 1.720624, "characteristics.90th_percentile_latency_ns": 1720624.0, "characteristics.90th_percentile_latency_s": 0.001720624, "characteristics.90th_percentile_latency_us": 1720.624, "ck_system": "DGX-A100_A100-SXM4x1_TRT", "ck_used": false, "cooling": "Air-cooled", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA V11.0.221", "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": 1024, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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-SXM4x1_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, DALI 0.25.0", "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": 1, "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-SXM4x1_TRT", "system_name": "NF5488A5", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 588.235, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "08dff605542a33c0", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 5087287, "90.00 percentile latency (ns)": 6245274, "90th percentile latency (ns)": 6245274, "95.00 percentile latency (ns)": 6473610, "97.00 percentile latency (ns)": 6577158, "99.00 percentile latency (ns)": 6713944, "99.90 percentile latency (ns)": 6908932, "Max latency (ns)": 7285149, "Mean latency (ns)": 5172420, "Min duration satisfied": "Yes", "Min latency (ns)": 4167396, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 193.14, "QPS w/o loadgen overhead": 193.33, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "accelerator_frequency": "1590MHz", "accelerator_host_interconnect": "PCIE", "accelerator_interconnect": "PCIE", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 6.245274, "characteristics.90th_percentile_latency_ns": 6245274.0, "characteristics.90th_percentile_latency_s": 0.006245274, "characteristics.90th_percentile_latency_us": 6245.274, "ck_system": "T4x1_TRT", "ck_used": false, "cooling": "Air-cooled", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.90th_percentile_latency_ms", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA V11.0.221", "host_memory_capacity": "1.5T", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "L1d cache: 32K, L1i cache: 32K, L2 cache: 1024K, L3 cache: 36608K", "host_processor_core_count": 24, "host_processor_frequency": "2.90GHz", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8268 CPU @ 2.90GHz", "host_processors_per_node": 2, "host_storage_capacity": "2.9 TB", "host_storage_type": "SSD", "hw_notes": "ECC off", "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": 1024, "mlperf_version": 0.7, "normalize_cores": 1, "normalize_processors": 1, "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/T4x1_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA V11.0.221, cuDNN 8.0.2, DALI 0.25.0", "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": 1, "schedule_rng_seed": 3135815929913719677, "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/T4x1_TRT", "system_name": "NF5468M5", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 156.25, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "4b3f8e005908df2a", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" } ]