[ { "50.00 percentile latency (ns)": 4685821, "90.00 percentile latency (ns)": 6014681, "90th percentile latency (ns)": 6014681, "95.00 percentile latency (ns)": 6404134, "97.00 percentile latency (ns)": 6761537, "99.00 percentile latency (ns)": 6971979, "99.90 percentile latency (ns)": 7243742, "Max latency (ns)": 7680805, "Mean latency (ns)": 4798528, "Min duration satisfied": "Yes", "Min latency (ns)": 3608900, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 208.11, "QPS w/o loadgen overhead": 208.4, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 6.014681, "characteristics.90th_percentile_latency_ns": 6014681.0, "characteristics.90th_percentile_latency_s": 0.006014681, "characteristics.90th_percentile_latency_us": 6014.681, "ck_system": "T4x1_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 8.0.2, CUDA 11.3", "host_memory_capacity": "128 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 4, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) D-2123IT CPU @ 2.20GHz", "host_processors_per_node": 1, "host_storage_capacity": "2 TB", "host_storage_type": "M.2", "hw_notes": "ECC on", "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": 1.1, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Lenovo/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Lenovo/results/T4x1_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02", "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": 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": "Lenovo", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Lenovo", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x1_TRT", "system_name": "Lenovo ThinkSystem SE350 server", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 156.25, "task": "NLP", "task2": "nlp", "total_cores": 4, "uid": "c7b14db19cd3a11b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 5284248, "90.00 percentile latency (ns)": 6588767, "90th percentile latency (ns)": 6588767, "95.00 percentile latency (ns)": 6828587, "97.00 percentile latency (ns)": 6950433, "99.00 percentile latency (ns)": 7110822, "99.90 percentile latency (ns)": 7386122, "Max latency (ns)": 7669486, "Mean latency (ns)": 5380097, "Min duration satisfied": "Yes", "Min latency (ns)": 4128910, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "QPS w/ loadgen overhead": 185.64, "QPS w/o loadgen overhead": 185.87, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 6.588767, "characteristics.90th_percentile_latency_ns": 6588767.0, "characteristics.90th_percentile_latency_s": 0.006588767, "characteristics.90th_percentile_latency_us": 6588.767, "ck_system": "T4x1_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 7.2.3, CUDA 11.1", "host_memory_capacity": "128 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 4, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) D-2123IT CPU @ 2.20GHz", "host_processors_per_node": 1, "host_storage_capacity": "2 TB", "host_storage_type": "M.2", "hw_notes": "ECC on", "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": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Lenovo/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Lenovo/results/T4x1_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "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": "Lenovo", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Lenovo", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x1_TRT", "system_name": "Lenovo ThinkSystem SE350 server", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 156.25, "task": "NLP", "task2": "nlp", "total_cores": 4, "uid": "85c8f8f674703a19", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 4953632, "90.00 percentile latency (ns)": 6246506, "90th percentile latency (ns)": 6246506, "95.00 percentile latency (ns)": 6461424, "97.00 percentile latency (ns)": 6586068, "99.00 percentile latency (ns)": 6729503, "99.90 percentile latency (ns)": 6956413, "Max latency (ns)": 7158262, "Mean latency (ns)": 5053263, "Min duration satisfied": "Yes", "Min latency (ns)": 3854009, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 197.62, "QPS w/o loadgen overhead": 197.89, "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "singlestream", "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": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 6.246506, "characteristics.90th_percentile_latency_ns": 6246506.0, "characteristics.90th_percentile_latency_s": 0.006246506, "characteristics.90th_percentile_latency_us": 6246.506, "ck_system": "T4x1_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 11.0 Update 1", "host_memory_capacity": "128 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 4, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) D-2123IT CPU @ 2.20GHz", "host_processors_per_node": 1, "host_storage_capacity": "2 TB", "host_storage_type": "M.2", "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/Lenovo/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Lenovo/results/T4x1_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, 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": "Lenovo", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Lenovo", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x1_TRT", "system_name": "LenovoSE350 (1x T4)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 156.25, "task": "NLP", "task2": "nlp", "total_cores": 4, "uid": "c86e9f8e273ff4a6", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" } ]