[ { "50.00 percentile latency (ns)": 447806748176, "90.00 percentile latency (ns)": 663538560624, "95.00 percentile latency (ns)": 682583207501, "97.00 percentile latency (ns)": 688934449834, "99.00 percentile latency (ns)": 694072606228, "99.90 percentile latency (ns)": 695851276183, "Max latency (ns)": 695948824434, "Mean latency (ns)": 420660668728, "Min duration satisfied": "Yes", "Min latency (ns)": 1133313399, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 1138.01, "Scenario": "offline", "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "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": "2.12.3", "characteristics.samples_per_second": 1138.01, "characteristics.samples_per_second.normalized_per_core": 1138.01, "characteristics.samples_per_second.normalized_per_processor": 1138.01, "ck_system": "XE2420_A10x1_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", "filesystem": "ext3/ext4", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "air", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6252N CPU @ 2.30GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "5.00.00.00", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE2420_A10x1_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, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "power_settings": "closed/Dell/power/XE2420_A10x1_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 792000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_A10x1_TRT", "system_name": "Dell EMC PowerEdge XE2420 (1x A10, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1200, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "2f7abc4074744749", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 449626164127, "90.00 percentile latency (ns)": 664110803260, "95.00 percentile latency (ns)": 683159067342, "97.00 percentile latency (ns)": 689389267808, "99.00 percentile latency (ns)": 694503275560, "99.90 percentile latency (ns)": 696278476480, "Max latency (ns)": 696394898954, "Mean latency (ns)": 422396131922, "Min duration satisfied": "Yes", "Min latency (ns)": 1280501119, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 1137.29, "Scenario": "offline", "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "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": "2.12.3", "characteristics.power": 409.6898134863705, "characteristics.power.normalized_per_core": 409.6898134863705, "characteristics.power.normalized_per_processor": 409.6898134863705, "characteristics.samples_per_second": 1137.29, "characteristics.samples_per_second.normalized_per_core": 1137.29, "characteristics.samples_per_second.normalized_per_processor": 1137.29, "ck_system": "XE2420_A10x1_TRT_MaxQ", "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", "filesystem": "ext3/ext4", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "air", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6252N CPU @ 2.30GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "5.00.00.00", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE2420_A10x1_TRT_MaxQ", "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, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "power_settings": "closed/Dell/power/XE2420_A10x1_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 792000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_A10x1_TRT_MaxQ", "system_name": "Dell EMC PowerEdge XE2420 (1x A10, MaxQ, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1200, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "1734dd092888abf6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 568700071454, "90.00 percentile latency (ns)": 840052481669, "95.00 percentile latency (ns)": 864097498520, "97.00 percentile latency (ns)": 872075335865, "99.00 percentile latency (ns)": 878530660640, "99.90 percentile latency (ns)": 880755525069, "Max latency (ns)": 880975955543, "Mean latency (ns)": 534339038140, "Min duration satisfied": "Yes", "Min latency (ns)": 1555760805, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT SERVER", "Samples per second": 2210.05, "Scenario": "offline", "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "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": 2210.05, "characteristics.samples_per_second.normalized_per_core": 1105.025, "characteristics.samples_per_second.normalized_per_processor": 1105.025, "ck_system": "XR12_edge_A10x2_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", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "host_memory_capacity": "512 GB", "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 6330 CPU @ 2.00GHz", "host_processors_per_node": 1, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "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, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XR12_edge_A10x2_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, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "power_settings": "", "print_timestamps": 0, "problem": false, "psu_details": "", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1947000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "bert_large_v1_1_fake_quant.onnx", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XR12_edge_A10x2_TRT", "system_name": "Dell EMC PowerEdge XR12 (2x A10, TensorRT)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 2950, "task": "NLP", "task2": "nlp", "total_cores": 28, "uid": "3546dba19ab6c202", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]