[ { "50.00 percentile latency (ns)": 398530697, "90.00 percentile latency (ns)": 488844260, "95.00 percentile latency (ns)": 545407149, "97.00 percentile latency (ns)": 554912776, "99.00 percentile latency (ns)": 568066721, "99.90 percentile latency (ns)": 635478003, "Completed samples per second": 108953.92, "Max latency (ns)": 734188896, "Mean latency (ns)": 386630544, "Min duration satisfied": "Yes", "Min latency (ns)": 38240726, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 109012.68, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 109012.68, "characteristics.scheduled_queries_per_second.normalized_per_core": 13626.585, "characteristics.scheduled_queries_per_second.normalized_per_processor": 13626.585, "characteristics.word error rate": 7.4454891880312974, "ck_system": "NF5688M6_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "LibriSpeech", "dataset_link": "", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 32, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8358", "host_processors_per_node": 2, "host_storage_capacity": "28 TB + 448GB*2", "host_storage_type": "NVMe SSD + M.2", "hw_notes": "NVIDIA A100-SXM-80GB-500W", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NF5688M6_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "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": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5688M6_A100-SXM-80GBx8_TRT", "system_name": "NF5688M6 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 109000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "d9b2d16c669d4aa9", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 402111278, "90.00 percentile latency (ns)": 493564157, "95.00 percentile latency (ns)": 549617925, "97.00 percentile latency (ns)": 560019607, "99.00 percentile latency (ns)": 573997118, "99.90 percentile latency (ns)": 641961218, "Completed samples per second": 107900.8, "Max latency (ns)": 774962147, "Mean latency (ns)": 391008242, "Min duration satisfied": "Yes", "Min latency (ns)": 51106502, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 107962.9, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 107962.9, "characteristics.scheduled_queries_per_second.normalized_per_core": 13495.3625, "characteristics.scheduled_queries_per_second.normalized_per_processor": 13495.3625, "characteristics.word error rate": 7.449998872578862, "ck_system": "NF5488A5_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "Liquid", "dataset": "LibriSpeech", "dataset_link": "", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.1, CUDA 11.3", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7742", "host_processors_per_node": 2, "host_storage_capacity": "11.7 TB + 448GB*2", "host_storage_type": "NVMe SSD + M.2", "hw_notes": "NVIDIA A100-SXM-80GB-500W", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "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/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Inspur/results/NF5488A5_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "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": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 107950, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "3e291a40398d15f0", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 336677335, "90.00 percentile latency (ns)": 458162558, "95.00 percentile latency (ns)": 486400206, "97.00 percentile latency (ns)": 518225769, "99.00 percentile latency (ns)": 548890529, "99.90 percentile latency (ns)": 629308232, "Completed samples per second": 104935.48, "Max latency (ns)": 865092941, "Mean latency (ns)": 347807891, "Min duration satisfied": "Yes", "Min latency (ns)": 82112614, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 104986.8, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 104986.8, "characteristics.scheduled_queries_per_second.normalized_per_core": 13123.35, "characteristics.scheduled_queries_per_second.normalized_per_processor": 13123.35, "characteristics.word error rate": 7.449998872578862, "ck_system": "NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "Liquid", "dataset": "LibriSpeech", "dataset_link": "", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.2", "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 7742", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.2", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 1, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "Preview", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_LIQUID_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5 (Liquid cooling)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 105000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "0cb6515d41edbba2", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 424790076, "90.00 percentile latency (ns)": 571548880, "95.00 percentile latency (ns)": 606171969, "97.00 percentile latency (ns)": 637156106, "99.00 percentile latency (ns)": 680162509, "99.90 percentile latency (ns)": 760014753, "Completed samples per second": 85449.36, "Max latency (ns)": 900913620, "Mean latency (ns)": 434086225, "Min duration satisfied": "Yes", "Min latency (ns)": 63009765, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 85485.6, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 85485.6, "characteristics.scheduled_queries_per_second.normalized_per_core": 10685.7, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10685.7, "characteristics.word error rate": 7.449998872578862, "ck_system": "NF5468M6_A100-PCIex8_TRT", "ck_used": true, "cooling": "", "dataset": "LibriSpeech", "dataset_link": "", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.1", "host_memory_capacity": "2 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "15 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5468M6_A100-PCIex8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.1, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 1, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "Preview", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5468M6_A100-PCIex8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 85500, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "4954330492905481", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 371067234, "90.00 percentile latency (ns)": 484856070, "95.00 percentile latency (ns)": 525006670, "97.00 percentile latency (ns)": 547671299, "99.00 percentile latency (ns)": 574908897, "99.90 percentile latency (ns)": 642673642, "Completed samples per second": 104834.63, "Max latency (ns)": 800540291, "Mean latency (ns)": 374334661, "Min duration satisfied": "Yes", "Min latency (ns)": 79399812, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 104886.02, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 104886.02, "characteristics.scheduled_queries_per_second.normalized_per_core": 13110.7525, "characteristics.scheduled_queries_per_second.normalized_per_processor": 13110.7525, "characteristics.word error rate": 7.4454891880312974, "ck_system": "NF5488A5_A100-SXM-80GBx8_TRT", "ck_used": false, "cooling": "", "dataset": "LibriSpeech", "dataset_link": "", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2, CUDA 11.2", "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 7742", "host_processors_per_node": 2, "host_storage_capacity": "12 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.0, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Inspur/results/NF5488A5_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.2", "other_software_stack": "TensorRT 7.2, CUDA 11.2, cuDNN 8.0.2, DALI 0.25.0, Triton 20.09", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "N", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 1, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Inspur", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Inspur", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/NF5488A5_A100-SXM-80GBx8_TRT", "system_name": "NF5488A5", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 104900, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "d86feda8e1b40419", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" } ]