[ { "50.00 percentile latency (ns)": 64291935, "90.00 percentile latency (ns)": 90945871, "95.00 percentile latency (ns)": 97833775, "97.00 percentile latency (ns)": 102715425, "99.00 percentile latency (ns)": 113476251, "99.90 percentile latency (ns)": 137036269, "Completed samples per second": 9985.02, "Max latency (ns)": 1024777779, "Mean latency (ns)": 65315136, "Min duration satisfied": "Yes", "Min latency (ns)": 18644940, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Scenario": "server", "Scheduled samples per second": 10001.94, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB (7x1g.10gb MIG)", "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.scheduled_queries_per_second": 10001.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 2500.485, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2500.485, "ck_system": "R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "open", "filesystem": "", "formal_model": "bert-99", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/open/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/open/Dell/results/R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 10000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "2642c5f288496abe", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 62691037, "90.00 percentile latency (ns)": 88709982, "95.00 percentile latency (ns)": 95049540, "97.00 percentile latency (ns)": 99252649, "99.00 percentile latency (ns)": 108463011, "99.90 percentile latency (ns)": 127541967, "Completed samples per second": 9985.15, "Max latency (ns)": 1025950343, "Mean latency (ns)": 63576614, "Min duration satisfied": "Yes", "Min latency (ns)": 19012047, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_MultiMigServer", "Scenario": "server", "Scheduled samples per second": 10001.94, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB (7x1g.10gb MIG)", "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.scheduled_queries_per_second": 10001.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 2500.485, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2500.485, "ck_system": "R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "disk_controllers": "", "disk_drives": "", "division": "open", "filesystem": "", "formal_model": "bert-99.9", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/open/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/open/Dell/results/R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GB-MIG_28x1g.10gb_TRT_Triton", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB-MIG-7x1g.10gb, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 10000, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "c8d841d49cea8bc7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 71293531, "90.00 percentile latency (ns)": 92192885, "95.00 percentile latency (ns)": 97656178, "97.00 percentile latency (ns)": 101177155, "99.00 percentile latency (ns)": 107531497, "99.90 percentile latency (ns)": 117530109, "Completed samples per second": 7871.17, "Max latency (ns)": 132030117, "Mean latency (ns)": 71061853, "Min duration satisfied": "Yes", "Min latency (ns)": 3839961, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 7871.94, "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": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 7871.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 2623.98, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2623.98, "ck_system": "R7525_A100-PCIE-40GBx3_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "512 GB", "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": "AMD EPYC 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "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/R7525_A100-PCIE-40GBx3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A100-PCIE-40GBx3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A100-PCIE-40GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 7870, "task": "NLP", "task2": "nlp", "total_cores": 64, "uid": "55e6381234fe3bee", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 62630888, "90.00 percentile latency (ns)": 96210920, "95.00 percentile latency (ns)": 106761402, "97.00 percentile latency (ns)": 113787614, "99.00 percentile latency (ns)": 127185096, "99.90 percentile latency (ns)": 154211250, "Completed samples per second": 3814.6, "Max latency (ns)": 192997192, "Mean latency (ns)": 64973681, "Min duration satisfied": "Yes", "Min latency (ns)": 4134622, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 3814.94, "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": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 3814.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 1271.6466666666668, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1271.6466666666668, "ck_system": "R7525_A100-PCIE-40GBx3_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "512 GB", "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": "AMD EPYC 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "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/R7525_A100-PCIE-40GBx3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A100-PCIE-40GBx3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A100-PCIE-40GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 3812, "task": "NLP", "task2": "nlp", "total_cores": 64, "uid": "634a44bf249e180d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 70316216, "90.00 percentile latency (ns)": 87321199, "95.00 percentile latency (ns)": 91314219, "97.00 percentile latency (ns)": 93687371, "99.00 percentile latency (ns)": 97935853, "99.90 percentile latency (ns)": 104795060, "Completed samples per second": 11700.14, "Max latency (ns)": 115825994, "Mean latency (ns)": 69636102, "Min duration satisfied": "Yes", "Min latency (ns)": 3765215, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 11701.09, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "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.scheduled_queries_per_second": 11701.09, "characteristics.scheduled_queries_per_second.normalized_per_core": 2925.2725, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2925.2725, "ck_system": "R750xa_A100-PCIE-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 11700, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "79ede446d59963a8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 54406192, "90.00 percentile latency (ns)": 82743871, "95.00 percentile latency (ns)": 91616239, "97.00 percentile latency (ns)": 97562775, "99.00 percentile latency (ns)": 108990237, "99.90 percentile latency (ns)": 128668087, "Completed samples per second": 5682.42, "Max latency (ns)": 173148840, "Mean latency (ns)": 56261914, "Min duration satisfied": "Yes", "Min latency (ns)": 3955315, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 5682.9, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "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.scheduled_queries_per_second": 5682.9, "characteristics.scheduled_queries_per_second.normalized_per_core": 1420.725, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1420.725, "ck_system": "R750xa_A100-PCIE-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 5680, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "a29a53bb144d2584", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 52311358, "90.00 percentile latency (ns)": 69793972, "95.00 percentile latency (ns)": 75107740, "97.00 percentile latency (ns)": 78884576, "99.00 percentile latency (ns)": 89936109, "99.90 percentile latency (ns)": 525887628980, "Completed samples per second": 1849.26, "Max latency (ns)": 604725715053, "Mean latency (ns)": 2364501132, "Min duration satisfied": "Yes", "Min latency (ns)": 4628320, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 1849.4, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.power": 575.9031074380163, "characteristics.power.normalized_per_core": 287.95155371900813, "characteristics.power.normalized_per_processor": 287.95155371900813, "characteristics.scheduled_queries_per_second": 1849.4, "characteristics.scheduled_queries_per_second.normalized_per_core": 924.7, "characteristics.scheduled_queries_per_second.normalized_per_processor": 924.7, "ck_system": "XE2420_A10x2_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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": "384 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 6252 CPU @ 2.10GHz", "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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/XE2420_A10x2_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_A10x2_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": 1, "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_A10x2_TRT_MaxQ", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 1850, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "91054c7b84a771fd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32811458, "90.00 percentile latency (ns)": 49135068, "95.00 percentile latency (ns)": 54498566, "97.00 percentile latency (ns)": 58401581, "99.00 percentile latency (ns)": 68363311, "99.90 percentile latency (ns)": 520053787894, "Completed samples per second": 839.88, "Max latency (ns)": 603882999992, "Mean latency (ns)": 2210618713, "Min duration satisfied": "Yes", "Min latency (ns)": 5114503, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 839.96, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.power": 583.2093884297519, "characteristics.power.normalized_per_core": 291.60469421487596, "characteristics.power.normalized_per_processor": 291.60469421487596, "characteristics.scheduled_queries_per_second": 839.96, "characteristics.scheduled_queries_per_second.normalized_per_core": 419.98, "characteristics.scheduled_queries_per_second.normalized_per_processor": 419.98, "ck_system": "XE2420_A10x2_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "air", "host_memory_capacity": "384 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 6252 CPU @ 2.10GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "5.00.00.00", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/XE2420_A10x2_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_A10x2_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": 1, "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_A10x2_TRT_MaxQ", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 840, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "4c07b6d3136c8401", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 88508865, "90.00 percentile latency (ns)": 98210642, "95.00 percentile latency (ns)": 101702942, "97.00 percentile latency (ns)": 104197605, "99.00 percentile latency (ns)": 109636786, "99.90 percentile latency (ns)": 117904366, "Completed samples per second": 30101.76, "Max latency (ns)": 126362885, "Mean latency (ns)": 86625291, "Min duration satisfied": "Yes", "Min latency (ns)": 4273792, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 30105.21, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 30105.21, "characteristics.scheduled_queries_per_second.normalized_per_core": 3010.5209999999997, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3010.5209999999997, "ck_system": "DSS8440_A100-PCIE-80GBx10_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": " 768GB", "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 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 10, "normalize_processors": 10, "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/DSS8440_A100-PCIE-80GBx10_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A100-PCIE-80GBx10_TRT", "system_name": "Dell EMC DSS 8440 (10x NVIDIA A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 30110, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "1e70489c8e7492d6", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 65952813, "90.00 percentile latency (ns)": 98075994, "95.00 percentile latency (ns)": 107620955, "97.00 percentile latency (ns)": 113828903, "99.00 percentile latency (ns)": 125979773, "99.90 percentile latency (ns)": 144188157, "Completed samples per second": 14537.95, "Max latency (ns)": 171586267, "Mean latency (ns)": 67329720, "Min duration satisfied": "Yes", "Min latency (ns)": 4183858, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 14539.86, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 14539.86, "characteristics.scheduled_queries_per_second.normalized_per_core": 1453.986, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1453.986, "ck_system": "DSS8440_A100-PCIE-80GBx10_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": " 768GB", "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 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 10, "normalize_processors": 10, "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/DSS8440_A100-PCIE-80GBx10_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A100-PCIE-80GBx10_TRT", "system_name": "Dell EMC DSS 8440 (10x NVIDIA A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 14540, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "5f844cc8f45aed61", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 25272959, "90.00 percentile latency (ns)": 28157489, "95.00 percentile latency (ns)": 28921243, "97.00 percentile latency (ns)": 29423462, "99.00 percentile latency (ns)": 30430775, "99.90 percentile latency (ns)": 32596523, "Completed samples per second": 10684.43, "Max latency (ns)": 1003836133, "Mean latency (ns)": 25322524, "Min duration satisfied": "Yes", "Min latency (ns)": 17950123, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "server", "Scheduled samples per second": 10702.3, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "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.scheduled_queries_per_second": 10702.3, "characteristics.scheduled_queries_per_second.normalized_per_core": 2675.575, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2675.575, "ck_system": "R750xa_A100-PCIE-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT_Triton", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 10700, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "ed7c317449c4509a", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 51936385, "90.00 percentile latency (ns)": 61086474, "95.00 percentile latency (ns)": 64736121, "97.00 percentile latency (ns)": 67444395, "99.00 percentile latency (ns)": 73445590, "99.90 percentile latency (ns)": 85634473, "Completed samples per second": 5673.36, "Max latency (ns)": 1010193203, "Mean latency (ns)": 52757927, "Min duration satisfied": "Yes", "Min latency (ns)": 35540559, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "server", "Scheduled samples per second": 5682.9, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "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.scheduled_queries_per_second": 5682.9, "characteristics.scheduled_queries_per_second.normalized_per_core": 1420.725, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1420.725, "ck_system": "R750xa_A100-PCIE-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT_Triton", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 5680, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "04d7dde6e4d54bbe", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 47421110, "90.00 percentile latency (ns)": 70437645, "95.00 percentile latency (ns)": 77905060, "97.00 percentile latency (ns)": 83191800, "99.00 percentile latency (ns)": 98228209, "99.90 percentile latency (ns)": 529143559242, "Completed samples per second": 11500.39, "Max latency (ns)": 604119503762, "Mean latency (ns)": 2279000376, "Min duration satisfied": "Yes", "Min latency (ns)": 4602312, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 11501.43, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 11501.43, "characteristics.scheduled_queries_per_second.normalized_per_core": 1437.67875, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1437.67875, "ck_system": "DSS8440_A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "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/DSS8440_A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 11500, "task": "NLP", "task2": "nlp", "total_cores": 96, "uid": "4d42b64b64fe2b70", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 45564485, "90.00 percentile latency (ns)": 72017673, "95.00 percentile latency (ns)": 81039942, "97.00 percentile latency (ns)": 87530118, "99.00 percentile latency (ns)": 107117729, "99.90 percentile latency (ns)": 518496846984, "Completed samples per second": 5252.4, "Max latency (ns)": 603709061971, "Mean latency (ns)": 2231764767, "Min duration satisfied": "Yes", "Min latency (ns)": 5478794, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 5252.86, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 5252.86, "characteristics.scheduled_queries_per_second.normalized_per_core": 656.6075, "characteristics.scheduled_queries_per_second.normalized_per_processor": 656.6075, "ck_system": "DSS8440_A30x8_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "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/DSS8440_A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 5250, "task": "NLP", "task2": "nlp", "total_cores": 96, "uid": "877e33499fbb997e", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 33601498, "90.00 percentile latency (ns)": 47176338, "95.00 percentile latency (ns)": 51580488, "97.00 percentile latency (ns)": 54759352, "99.00 percentile latency (ns)": 65008940, "99.90 percentile latency (ns)": 523740978212, "Completed samples per second": 3962.03, "Max latency (ns)": 604144536883, "Mean latency (ns)": 2523178139, "Min duration satisfied": "Yes", "Min latency (ns)": 4980738, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 3962.46, "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": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 3962.46, "characteristics.scheduled_queries_per_second.normalized_per_core": 1320.82, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1320.82, "ck_system": "R7525_A30x3_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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 7763", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "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/R7525_A30x3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.4.2105", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A30x3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 3960, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "602061a04ebf921b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 47150053, "90.00 percentile latency (ns)": 73212610, "95.00 percentile latency (ns)": 81397863, "97.00 percentile latency (ns)": 87347928, "99.00 percentile latency (ns)": 108768653, "99.90 percentile latency (ns)": 528708960251, "Completed samples per second": 1949.8, "Max latency (ns)": 604021013265, "Mean latency (ns)": 2555961939, "Min duration satisfied": "Yes", "Min latency (ns)": 5747642, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 1949.88, "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": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 1949.88, "characteristics.scheduled_queries_per_second.normalized_per_core": 649.96, "characteristics.scheduled_queries_per_second.normalized_per_processor": 649.96, "ck_system": "R7525_A30x3_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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 7763", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "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/R7525_A30x3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.4.2105", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A30x3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 1950, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "9b90b57614eb48aa", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 66823035, "90.00 percentile latency (ns)": 86015858, "95.00 percentile latency (ns)": 90986540, "97.00 percentile latency (ns)": 94118867, "99.00 percentile latency (ns)": 99737500, "99.90 percentile latency (ns)": 108636542, "Completed samples per second": 7800.84, "Max latency (ns)": 122959662, "Mean latency (ns)": 66525410, "Min duration satisfied": "Yes", "Min latency (ns)": 8357397, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 7801.53, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 7801.53, "characteristics.scheduled_queries_per_second.normalized_per_core": 2600.5099999999998, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2600.5099999999998, "ck_system": "R7525_vA100-PCIE-40GBx3_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "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/R7525_vA100-PCIE-40GBx3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "VMware Submission", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_vA100-PCIE-40GBx3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x GRID A100-40C, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 7800, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "1f0459015a04a839", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 39735573, "90.00 percentile latency (ns)": 60423352, "95.00 percentile latency (ns)": 67279242, "97.00 percentile latency (ns)": 72087978, "99.00 percentile latency (ns)": 82080334, "99.90 percentile latency (ns)": 100399717, "Completed samples per second": 3601.91, "Max latency (ns)": 124926508, "Mean latency (ns)": 41530372, "Min duration satisfied": "Yes", "Min latency (ns)": 3713171, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 3602.11, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 3602.11, "characteristics.scheduled_queries_per_second.normalized_per_core": 1200.7033333333334, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1200.7033333333334, "ck_system": "R7525_vA100-PCIE-40GBx3_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 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 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 3, "normalize_processors": 3, "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/R7525_vA100-PCIE-40GBx3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "VMware Submission", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_vA100-PCIE-40GBx3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x GRID A100-40C, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 3600, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "389049f3874283e5", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 63389773, "90.00 percentile latency (ns)": 95412837, "95.00 percentile latency (ns)": 104637541, "97.00 percentile latency (ns)": 110495125, "99.00 percentile latency (ns)": 121160556, "99.90 percentile latency (ns)": 137457581, "Completed samples per second": 7800.89, "Max latency (ns)": 153931697, "Mean latency (ns)": 65390188, "Min duration satisfied": "Yes", "Min latency (ns)": 7063083, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 7801.53, "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.power": 1480.3514143094853, "characteristics.power.normalized_per_core": 370.08785357737133, "characteristics.power.normalized_per_processor": 370.08785357737133, "characteristics.scheduled_queries_per_second": 7801.53, "characteristics.scheduled_queries_per_second.normalized_per_core": 1950.3825, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1950.3825, "ck_system": "R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "256 GB", "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 8368 CPU @ 2.40GHz", "host_processors_per_node": 2, "host_storage_capacity": "543 GB", "host_storage_type": "SSD", "hw_notes": "Result Measured for Power; GPU Power Limited to 175W", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "power_management": "", "power_settings": "closed/Dell/power/R750xa_A100-PCIE-40GBx4_TRT_MaxQ.md", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "1x2400W", "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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-40GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 7800, "task": "NLP", "task2": "nlp", "total_cores": 64, "uid": "9484c42a2102a661", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 38969342, "90.00 percentile latency (ns)": 66029902, "95.00 percentile latency (ns)": 76617775, "97.00 percentile latency (ns)": 84188817, "99.00 percentile latency (ns)": 101087927, "99.90 percentile latency (ns)": 136299538, "Completed samples per second": 3551.57, "Max latency (ns)": 184708896, "Mean latency (ns)": 42487815, "Min duration satisfied": "Yes", "Min latency (ns)": 4457785, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 3551.72, "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.power": 1468.5383333333336, "characteristics.power.normalized_per_core": 367.1345833333334, "characteristics.power.normalized_per_processor": 367.1345833333334, "characteristics.scheduled_queries_per_second": 3551.72, "characteristics.scheduled_queries_per_second.normalized_per_core": 887.93, "characteristics.scheduled_queries_per_second.normalized_per_processor": 887.93, "ck_system": "R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "256 GB", "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 8368 CPU @ 2.40GHz", "host_processors_per_node": 2, "host_storage_capacity": "543 GB", "host_storage_type": "SSD", "hw_notes": "Result Measured for Power; GPU Power Limited to 175W", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "power_management": "", "power_settings": "closed/Dell/power/R750xa_A100-PCIE-40GBx4_TRT_MaxQ.md", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "1x2400W", "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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-40GBx4_TRT_MaxQ", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-40GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 3550, "task": "NLP", "task2": "nlp", "total_cores": 64, "uid": "93fd738f7b40218d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 59898136, "90.00 percentile latency (ns)": 79347045, "95.00 percentile latency (ns)": 86081095, "97.00 percentile latency (ns)": 91424326, "99.00 percentile latency (ns)": 107472592, "99.90 percentile latency (ns)": 526429959177, "Completed samples per second": 1949.7, "Max latency (ns)": 604678835154, "Mean latency (ns)": 2299372247, "Min duration satisfied": "Yes", "Min latency (ns)": 4648634, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 1949.88, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.scheduled_queries_per_second": 1949.88, "characteristics.scheduled_queries_per_second.normalized_per_core": 974.94, "characteristics.scheduled_queries_per_second.normalized_per_processor": 974.94, "ck_system": "XE2420_A10x2_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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": "384 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 6252 CPU @ 2.10GHz", "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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/XE2420_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": "closed/Dell/power/XE2420_A10x2_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": 1, "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_A10x2_TRT", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 1950, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "cae894e1586ec4e5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 49392033, "90.00 percentile latency (ns)": 71982132, "95.00 percentile latency (ns)": 79018988, "97.00 percentile latency (ns)": 84045360, "99.00 percentile latency (ns)": 98448322, "99.90 percentile latency (ns)": 523274890590, "Completed samples per second": 939.65, "Max latency (ns)": 604507345385, "Mean latency (ns)": 2286692869, "Min duration satisfied": "Yes", "Min latency (ns)": 5185766, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 939.74, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.scheduled_queries_per_second": 939.74, "characteristics.scheduled_queries_per_second.normalized_per_core": 469.87, "characteristics.scheduled_queries_per_second.normalized_per_processor": 469.87, "ck_system": "XE2420_A10x2_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "air", "host_memory_capacity": "384 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 6252 CPU @ 2.10GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "5.00.00.00", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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/XE2420_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": "closed/Dell/power/XE2420_A10x2_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": 1, "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_A10x2_TRT", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 940, "task": "NLP", "task2": "nlp", "total_cores": 48, "uid": "49718fe12b38453b", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 25085622, "90.00 percentile latency (ns)": 44274974, "95.00 percentile latency (ns)": 46301902, "97.00 percentile latency (ns)": 47385041, "99.00 percentile latency (ns)": 49218454, "99.90 percentile latency (ns)": 52093753, "Completed samples per second": 11001.14, "Max latency (ns)": 56783405, "Mean latency (ns)": 26393681, "Min duration satisfied": "Yes", "Min latency (ns)": 4875441, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "server", "Scheduled samples per second": 11001.55, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 11001.55, "characteristics.scheduled_queries_per_second.normalized_per_core": 1375.19375, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1375.19375, "ck_system": "DSS8440_A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "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/DSS8440_A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT_Triton", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 11000, "task": "NLP", "task2": "nlp", "total_cores": 96, "uid": "a107d6a41ebe4d7b", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 91191708, "90.00 percentile latency (ns)": 98320950, "95.00 percentile latency (ns)": 100248661, "97.00 percentile latency (ns)": 101509438, "99.00 percentile latency (ns)": 104021958, "99.90 percentile latency (ns)": 108401770, "Completed samples per second": 5193.88, "Max latency (ns)": 1054171786, "Mean latency (ns)": 91230949, "Min duration satisfied": "Yes", "Min latency (ns)": 69342737, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "server", "Scheduled samples per second": 5202.95, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.scheduled_queries_per_second": 5202.95, "characteristics.scheduled_queries_per_second.normalized_per_core": 650.36875, "characteristics.scheduled_queries_per_second.normalized_per_processor": 650.36875, "ck_system": "DSS8440_A30x8_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 8, "normalize_processors": 8, "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/DSS8440_A30x8_TRT_Triton", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, 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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT_Triton", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 5200, "task": "NLP", "task2": "nlp", "total_cores": 96, "uid": "694139d83d490cc3", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 51096716, "90.00 percentile latency (ns)": 67924475, "95.00 percentile latency (ns)": 72895741, "97.00 percentile latency (ns)": 76263583, "99.00 percentile latency (ns)": 85785994, "99.90 percentile latency (ns)": 523597781104, "Completed samples per second": 1849.3, "Max latency (ns)": 604365920544, "Mean latency (ns)": 2245331913, "Min duration satisfied": "Yes", "Min latency (ns)": 4505543, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 1849.4, "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.scheduled_queries_per_second": 1849.4, "characteristics.scheduled_queries_per_second.normalized_per_core": 924.7, "characteristics.scheduled_queries_per_second.normalized_per_processor": 924.7, "ck_system": "XR12_datacenter_A10x2_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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_datacenter_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": 1, "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_datacenter_A10x2_TRT", "system_name": "Dell EMC PowerEdge XR12 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 1850, "task": "NLP", "task2": "nlp", "total_cores": 28, "uid": "171b37238a499bba", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 32256124, "90.00 percentile latency (ns)": 48066811, "95.00 percentile latency (ns)": 53160450, "97.00 percentile latency (ns)": 56854434, "99.00 percentile latency (ns)": 66583382, "99.90 percentile latency (ns)": 514777335830, "Completed samples per second": 839.89, "Max latency (ns)": 603343368170, "Mean latency (ns)": 2126932883, "Min duration satisfied": "Yes", "Min latency (ns)": 5187928, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 839.96, "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.scheduled_queries_per_second": 839.96, "characteristics.scheduled_queries_per_second.normalized_per_core": 419.98, "characteristics.scheduled_queries_per_second.normalized_per_processor": 419.98, "ck_system": "XR12_datacenter_A10x2_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "division": "closed", "filesystem": "", "formal_model": "bert", "formal_model_accuracy": 99.9, "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-99.9", "input_data_types": "int32", "management_firmware_version": "", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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_datacenter_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": 1, "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_datacenter_A10x2_TRT", "system_name": "Dell EMC PowerEdge XR12 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 840, "task": "NLP", "task2": "nlp", "total_cores": 28, "uid": "5a344b7581715887", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 37890201, "90.00 percentile latency (ns)": 47348304, "95.00 percentile latency (ns)": 50935886, "97.00 percentile latency (ns)": 53415004, "99.00 percentile latency (ns)": 58099511, "99.90 percentile latency (ns)": 65088037, "Completed samples per second": 12905.98, "Max latency (ns)": 1014045620, "Mean latency (ns)": 39100689, "Min duration satisfied": "Yes", "Min latency (ns)": 26100422, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "server", "Scheduled samples per second": 12927.72, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.scheduled_queries_per_second": 12927.72, "characteristics.scheduled_queries_per_second.normalized_per_core": 3231.93, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3231.93, "ck_system": "XE8545_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "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_cooling_type": "", "host_memory_capacity": "1 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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "bert-99", "input_data_types": "int32", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "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/XE8545_A100-SXM-80GBx4_TRT_Triton", "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_management": "", "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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE8545_A100-SXM-80GBx4_TRT_Triton", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 12926, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "d6545c0bc37a90d5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 42430248, "90.00 percentile latency (ns)": 49114940, "95.00 percentile latency (ns)": 51756842, "97.00 percentile latency (ns)": 53688357, "99.00 percentile latency (ns)": 57946051, "99.90 percentile latency (ns)": 67958067, "Completed samples per second": 6390.81, "Max latency (ns)": 1034180078, "Mean latency (ns)": 42935280, "Min duration satisfied": "Yes", "Min latency (ns)": 28769317, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "Triton_Server", "Scenario": "server", "Scheduled samples per second": 6401.69, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.scheduled_queries_per_second": 6401.69, "characteristics.scheduled_queries_per_second.normalized_per_core": 1600.4225, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1600.4225, "ck_system": "XE8545_A100-SXM-80GBx4_TRT_Triton", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "host_memory_capacity": "1 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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "bert-99.9", "input_data_types": "int32", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "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/XE8545_A100-SXM-80GBx4_TRT_Triton", "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_management": "", "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": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE8545_A100-SXM-80GBx4_TRT_Triton", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT, Triton)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 6400, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "8e01a5ecff4a989c", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 58619850, "90.00 percentile latency (ns)": 67084445, "95.00 percentile latency (ns)": 70162212, "97.00 percentile latency (ns)": 72629582, "99.00 percentile latency (ns)": 89091485, "99.90 percentile latency (ns)": 545916104106, "Completed samples per second": 13790.63, "Max latency (ns)": 606525496737, "Mean latency (ns)": 3074482424, "Min duration satisfied": "Yes", "Min latency (ns)": 3836578, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 13791.67, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "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": "2.2.5", "characteristics.power": 3245.7612850082396, "characteristics.power.normalized_per_core": 811.4403212520599, "characteristics.power.normalized_per_processor": 811.4403212520599, "characteristics.scheduled_queries_per_second": 13791.67, "characteristics.scheduled_queries_per_second.normalized_per_core": 3447.9175, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3447.9175, "ck_system": "XE8545_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_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": "", "host_memory_capacity": "1 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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "bert-99", "input_data_types": "int32", "management_firmware_version": "4.40.40.151", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "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/XE8545_A100-SXM-80GBx4_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_management": "", "power_settings": "closed/Dell/power/XE8545_A100-SXM-80GBx4_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "4x2400W", "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": "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/XE8545_A100-SXM-80GBx4_TRT", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 13790, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "220a65d6cd83acb7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 47449275, "90.00 percentile latency (ns)": 58651482, "95.00 percentile latency (ns)": 61405258, "97.00 percentile latency (ns)": 63430451, "99.00 percentile latency (ns)": 75175733, "99.90 percentile latency (ns)": 544381073819, "Completed samples per second": 6901.69, "Max latency (ns)": 606801462819, "Mean latency (ns)": 3057311248, "Min duration satisfied": "Yes", "Min latency (ns)": 4363421, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "BERT SERVER", "Scenario": "server", "Scheduled samples per second": 6902.14, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "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": "2.2.5", "characteristics.power": 3139.3082372322874, "characteristics.power.normalized_per_core": 784.8270593080719, "characteristics.power.normalized_per_processor": 784.8270593080719, "characteristics.scheduled_queries_per_second": 6902.14, "characteristics.scheduled_queries_per_second.normalized_per_core": 1725.535, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1725.535, "ck_system": "XE8545_A100-SXM-80GBx4_TRT", "ck_used": false, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_y_default": "characteristics.scheduled_queries_per_second", "dim_y_maximize": false, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "host_memory_capacity": "1 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 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "bert-99.9", "input_data_types": "int32", "management_firmware_version": "4.40.40.151", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "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/XE8545_A100-SXM-80GBx4_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_management": "", "power_settings": "closed/Dell/power/XE8545_A100-SXM-80GBx4_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "4x2400W", "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": "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/XE8545_A100-SXM-80GBx4_TRT", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 130000000, "target_qps": 6900, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "3b504f2487db81e0", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" } ]