[ { "50.00 percentile latency (ns)": 44830104263, "90.00 percentile latency (ns)": 66268719631, "95.00 percentile latency (ns)": 68203959210, "97.00 percentile latency (ns)": 68765588204, "99.00 percentile latency (ns)": 69114090013, "99.90 percentile latency (ns)": 69199604611, "Max latency (ns)": 69199636337, "Mean latency (ns)": 42309489639, "Min duration satisfied": "Yes", "Min latency (ns)": 2604902214, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 1621.4, "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 T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 1621.4, "characteristics.samples_per_second.normalized_per_core": 405.35, "characteristics.samples_per_second.normalized_per_processor": 405.35, "ck_system": "T4x4_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 11.0 Update 1", "host_memory_capacity": "256 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 TB", "host_storage_type": "SATA HDD", "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": 1, "mlperf_version": 0.7, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Altos/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Altos/results/T4x4_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 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "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": 112200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "ALTOS", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Altos", "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/T4x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 1700, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "532e27a3640de92f", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 52891903383, "90.00 percentile latency (ns)": 77078572043, "95.00 percentile latency (ns)": 78998268744, "97.00 percentile latency (ns)": 80030209007, "99.00 percentile latency (ns)": 80239434465, "99.90 percentile latency (ns)": 80528634947, "Max latency (ns)": 80528648306, "Mean latency (ns)": 49426358009, "Min duration satisfied": "Yes", "Min latency (ns)": 3194958718, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 668.781, "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 T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 668.781, "characteristics.samples_per_second.normalized_per_core": 167.19525, "characteristics.samples_per_second.normalized_per_processor": 167.19525, "ck_system": "T4x4_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 11.0 Update 1", "host_memory_capacity": "256 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 TB", "host_storage_type": "SATA HDD", "hw_notes": "ECC off", "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": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Altos/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Altos/results/T4x4_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 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "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": 53856, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "ALTOS", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Altos", "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/T4x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 816, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "aa393fb56fc40c0b", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" } ]