[ { "50.00 percentile latency (ns)": 472010505, "90.00 percentile latency (ns)": 663455382, "95.00 percentile latency (ns)": 713570649, "97.00 percentile latency (ns)": 745203249, "99.00 percentile latency (ns)": 790713682, "99.90 percentile latency (ns)": 849351343, "Completed samples per second": 4199.94, "Max latency (ns)": 968301607, "Mean latency (ns)": 490871389, "Min duration satisfied": "Yes", "Min latency (ns)": 82665662, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 4202.02, "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "1.1.3", "characteristics.accuracy": 92.56127533879005, "characteristics.power": 862.4563333333342, "characteristics.power.normalized_per_core": 215.61408333333355, "characteristics.power.normalized_per_processor": 215.61408333333355, "characteristics.scheduled_queries_per_second": 4202.02, "characteristics.scheduled_queries_per_second.normalized_per_core": 1050.505, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1050.505, "characteristics.word error rate": 7.438724661209949, "ck_system": "XE2420_datacenter_T4x4_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", "filesystem": "ext3/ext4", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_cooling_type": "passive", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "4.40.10.00", "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": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/XE2420_datacenter_T4x4_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_settings": "closed/DellEMC/measurements/XE2420_datacenter_T4x4_TRT/XE2420_T4x4_power_settings.adoc", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_datacenter_T4x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 4200, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "b637640738b2777f", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 506656403, "90.00 percentile latency (ns)": 639863784, "95.00 percentile latency (ns)": 687045459, "97.00 percentile latency (ns)": 705883583, "99.00 percentile latency (ns)": 761207069, "99.90 percentile latency (ns)": 836286701, "Completed samples per second": 107928.96, "Max latency (ns)": 1035216483, "Mean latency (ns)": 507033339, "Min duration satisfied": "Yes", "Min latency (ns)": 65395361, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 107987.47, "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": 10, "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": 107987.47, "characteristics.scheduled_queries_per_second.normalized_per_core": 10798.747, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10798.747, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DSS8440_A100-PCIE-40GBx10_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, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "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": 1.0, "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_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/DSS8440_A100-PCIE-40GBx10_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_A100-PCIE-40GBx10_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 108000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "96a642aaa5ba08fb", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 364651821, "90.00 percentile latency (ns)": 499621410, "95.00 percentile latency (ns)": 536307205, "97.00 percentile latency (ns)": 559465777, "99.00 percentile latency (ns)": 599589383, "99.90 percentile latency (ns)": 664447857, "Completed samples per second": 51969.65, "Max latency (ns)": 766499149, "Mean latency (ns)": 375630625, "Min duration satisfied": "Yes", "Min latency (ns)": 45830144, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 51985.94, "accelerator_cooling_type": "", "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": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55225596969493, "characteristics.scheduled_queries_per_second": 51985.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 12996.485, "characteristics.scheduled_queries_per_second.normalized_per_processor": 12996.485, "characteristics.word error rate": 7.4477440303050795, "ck_system": "XE8545_7763_A100-SXM4-80GBx4_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 7.2.3, CUDA 11.1", "host_cooling_type": "", "host_memory_capacity": "1 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": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "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": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/XE8545_7763_A100-SXM4-80GBx4_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 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_management": "", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/XE8545_7763_A100-SXM4-80GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 52000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "eb56b77f9ed4edfd", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 416768636, "90.00 percentile latency (ns)": 540173914, "95.00 percentile latency (ns)": 581471747, "97.00 percentile latency (ns)": 608027240, "99.00 percentile latency (ns)": 631352295, "99.90 percentile latency (ns)": 703281295, "Completed samples per second": 49657.65, "Max latency (ns)": 877696224, "Mean latency (ns)": 422862351, "Min duration satisfied": "Yes", "Min latency (ns)": 46763687, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 49684.13, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 49684.13, "characteristics.scheduled_queries_per_second.normalized_per_core": 12421.0325, "characteristics.scheduled_queries_per_second.normalized_per_processor": 12421.0325, "characteristics.word error rate": 7.449998872578862, "ck_system": "XE8545_7713_A100-SXM4-40GBx4", "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 7.2.3, CUDA 11.1", "host_cooling_type": "", "host_memory_capacity": "1 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": "3 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": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/XE8545_7713_A100-SXM4-40GBx4", "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 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_management": "", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/XE8545_7713_A100-SXM4-40GBx4", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 49700, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "88030b54e9f8ba48", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 404932012, "90.00 percentile latency (ns)": 537039669, "95.00 percentile latency (ns)": 578903290, "97.00 percentile latency (ns)": 604933647, "99.00 percentile latency (ns)": 629241674, "99.90 percentile latency (ns)": 702457106, "Completed samples per second": 49660.83, "Max latency (ns)": 811763338, "Mean latency (ns)": 416907379, "Min duration satisfied": "Yes", "Min latency (ns)": 33101597, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 49684.13, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 49684.13, "characteristics.scheduled_queries_per_second.normalized_per_core": 12421.0325, "characteristics.scheduled_queries_per_second.normalized_per_processor": 12421.0325, "characteristics.word error rate": 7.449998872578862, "ck_system": "XE8545_7763_A100-SXM4-40GBx4_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 7.2.3, CUDA 11.1", "host_cooling_type": "", "host_memory_capacity": "1 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": "3 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": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/XE8545_7763_A100-SXM4-40GBx4_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 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_management": "", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/XE8545_7763_A100-SXM4-40GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 49700, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "d406829ed4282b7f", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 444756554, "90.00 percentile latency (ns)": 594876832, "95.00 percentile latency (ns)": 632093586, "97.00 percentile latency (ns)": 666133465, "99.00 percentile latency (ns)": 713246144, "99.90 percentile latency (ns)": 794559873, "Completed samples per second": 43960.33, "Max latency (ns)": 953404504, "Mean latency (ns)": 450184976, "Min duration satisfied": "Yes", "Min latency (ns)": 43162553, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 43986.32, "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.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 43986.32, "characteristics.scheduled_queries_per_second.normalized_per_core": 10996.58, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10996.58, "characteristics.word error rate": 7.449998872578862, "ck_system": "R750xa_A100-PCIE-40GBx4_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, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "256 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "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": 1.0, "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_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/R750xa_A100-PCIE-40GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/R750xa_A100-PCIE-40GBx4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 44000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "f8be57550d606664", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 471673470, "90.00 percentile latency (ns)": 715485029, "95.00 percentile latency (ns)": 803955049, "97.00 percentile latency (ns)": 868979468, "99.00 percentile latency (ns)": 996924772, "99.90 percentile latency (ns)": 1217298491, "Completed samples per second": 37371.29, "Max latency (ns)": 1660998823, "Mean latency (ns)": 495361245, "Min duration satisfied": "Yes", "Min latency (ns)": 74388389, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 37391.11, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "48 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A40", "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.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 37391.11, "characteristics.scheduled_queries_per_second.normalized_per_core": 3739.111, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3739.111, "characteristics.word error rate": 7.449998872578862, "ck_system": "DSS8440_A40x10_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, "disk_controllers": "", "disk_drives": "", "division": "closed", "filesystem": "", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 7.2.3, CUDA 11.1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "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": 1.0, "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_v1.0/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/DellEMC/results/DSS8440_A40x10_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0; 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, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "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": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_A40x10_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 37400, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "24c863c84ea4b25a", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 484014562, "90.00 percentile latency (ns)": 725791505, "95.00 percentile latency (ns)": 805240267, "97.00 percentile latency (ns)": 849059677, "99.00 percentile latency (ns)": 943805736, "99.90 percentile latency (ns)": 1106064601, "Completed samples per second": 8637.02, "Max latency (ns)": 1361505586, "Mean latency (ns)": 510107611, "Min duration satisfied": "Yes", "Min latency (ns)": 57691266, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 8697.69, "accelerator_frequency": "1590MHz", "accelerator_host_interconnect": "PCI Express 3.0", "accelerator_interconnect": "PCI Express 3.0", "accelerator_interconnect_topology": "4 Accelerators per CPU", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Tesla T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56127533879005, "characteristics.scheduled_queries_per_second": 8697.69, "characteristics.scheduled_queries_per_second.normalized_per_core": 1087.21125, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1087.21125, "characteristics.word error rate": 7.438724661209949, "ck_system": "R7525_T4x8_TRT", "ck_used": true, "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 11.0, cuDNN 8.0.2, cuBLAS 11.2.0, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "1 TB", "host_memory_configuration": "8x64GB DDR4-3200 HMAA8GR7AJR4N-XN RDIMM ECC", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "2MB+16MB+128MB", "host_processor_core_count": 120, "host_processor_frequency": "2.35GHz", "host_processor_interconnect": "Infinity Fabric", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "3 TB (5x800GB WUSTR6480ASS200 in RAID5)", "host_storage_type": "3D-TLC Solid State with 12Gbps SAS", "hw_notes": "ECC off", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/R7525_T4x8_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": "docker 19.03.12, python 3.6.8, gcc 5.5.0, onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/R7525_T4x8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 8700, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "0bce1464b44846f1", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 478650665, "90.00 percentile latency (ns)": 716758721, "95.00 percentile latency (ns)": 789404940, "97.00 percentile latency (ns)": 834125807, "99.00 percentile latency (ns)": 926149683, "99.90 percentile latency (ns)": 1084642755, "Completed samples per second": 4562.13, "Max latency (ns)": 1244192151, "Mean latency (ns)": 506615962, "Min duration satisfied": "Yes", "Min latency (ns)": 113861914, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 4584.87, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56353018106384, "characteristics.scheduled_queries_per_second": 4584.87, "characteristics.scheduled_queries_per_second.normalized_per_core": 1146.2175, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1146.2175, "characteristics.word error rate": 7.436469818936166, "ck_system": "XE2420_T4x4_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 11.0 Update 1", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "2.10GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/XE2420_T4x4_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/XE2420_T4x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 4590, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "0ab4ae9620944b0b", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 554794591, "90.00 percentile latency (ns)": 719205707, "95.00 percentile latency (ns)": 765590639, "97.00 percentile latency (ns)": 797640520, "99.00 percentile latency (ns)": 850134454, "99.90 percentile latency (ns)": 938083978, "Completed samples per second": 24440.32, "Max latency (ns)": 1032109926, "Mean latency (ns)": 564221210, "Min duration satisfied": "Yes", "Min latency (ns)": 60164054, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 24628.16, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "48 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Quadro RTX 8000", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56353018106384, "characteristics.scheduled_queries_per_second": 24628.16, "characteristics.scheduled_queries_per_second.normalized_per_core": 3078.52, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3078.52, "characteristics.word error rate": 7.436469818936166, "ck_system": "DSS8440_QuadroRTX8000x8_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 11.0 Update 1", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/DSS8440_QuadroRTX8000x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_QuadroRTX8000x8_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 24620, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "44c4bad816ec3595", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 483442337, "90.00 percentile latency (ns)": 621320489, "95.00 percentile latency (ns)": 663235072, "97.00 percentile latency (ns)": 687124866, "99.00 percentile latency (ns)": 729220480, "99.90 percentile latency (ns)": 809280845, "Completed samples per second": 29397.9, "Max latency (ns)": 954348563, "Mean latency (ns)": 488970151, "Min duration satisfied": "Yes", "Min latency (ns)": 58389511, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 29607.65, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Quadro RTX 6000", "accelerator_on-chip_memories": "", "accelerators_per_node": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56127533879005, "characteristics.scheduled_queries_per_second": 29607.65, "characteristics.scheduled_queries_per_second.normalized_per_core": 2960.7650000000003, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2960.7650000000003, "characteristics.word error rate": 7.438724661209949, "ck_system": "DSS8440_QuadroRTX6000x10_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 11.0 Update 1", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 10, "normalize_processors": 10, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/DSS8440_QuadroRTX6000x10_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_QuadroRTX6000x10_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 29600, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "9ca5553289f24485", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 565984407, "90.00 percentile latency (ns)": 799289863, "95.00 percentile latency (ns)": 841935359, "97.00 percentile latency (ns)": 887020516, "99.00 percentile latency (ns)": 961157592, "99.90 percentile latency (ns)": 1084148184, "Completed samples per second": 14504.13, "Max latency (ns)": 1249773513, "Mean latency (ns)": 583533177, "Min duration satisfied": "Yes", "Min latency (ns)": 46254945, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 14600.4, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56127533879005, "characteristics.scheduled_queries_per_second": 14600.4, "characteristics.scheduled_queries_per_second.normalized_per_core": 7300.2, "characteristics.scheduled_queries_per_second.normalized_per_processor": 7300.2, "characteristics.word error rate": 7.438724661209949, "ck_system": "R7525_A100x2_TRT", "ck_used": true, "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 11.0, cuDNN 8.0.2, cuBLAS 11.2.0, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "2.50GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.84 TB", "host_storage_type": "NVMe", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/R7525_A100x2_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": "docker 19.03.12, python 3.6.8, gcc 5.5.0, onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/R7525_A100x2_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 14600, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "4de4189b173144f5", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 570679847, "90.00 percentile latency (ns)": 803792199, "95.00 percentile latency (ns)": 846821493, "97.00 percentile latency (ns)": 898380252, "99.00 percentile latency (ns)": 961701575, "99.90 percentile latency (ns)": 1083110721, "Completed samples per second": 21873.61, "Max latency (ns)": 1277041468, "Mean latency (ns)": 591865269, "Min duration satisfied": "Yes", "Min latency (ns)": 38000158, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 22007.12, "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, "characteristics.accuracy": 92.5567656542425, "characteristics.scheduled_queries_per_second": 22007.12, "characteristics.scheduled_queries_per_second.normalized_per_core": 7335.706666666666, "characteristics.scheduled_queries_per_second.normalized_per_processor": 7335.706666666666, "characteristics.word error rate": 7.4432343457575145, "ck_system": "R7525_A100x3_TRT", "ck_used": true, "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 11.0, cuDNN 8.0.2, cuBLAS 11.2.0, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "2.40GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.5 TB", "host_storage_type": "NVMe", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/R7525_A100x3_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": "docker 19.03.12, python 3.6.8, gcc 5.5.0, onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "preview", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/R7525_A100x3_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 22000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "0cd3f4def93c3599", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 510494023, "90.00 percentile latency (ns)": 649577815, "95.00 percentile latency (ns)": 696884314, "97.00 percentile latency (ns)": 720955966, "99.00 percentile latency (ns)": 763304961, "99.90 percentile latency (ns)": 861870808, "Completed samples per second": 29299.03, "Max latency (ns)": 922969750, "Mean latency (ns)": 513964608, "Min duration satisfied": "Yes", "Min latency (ns)": 27742613, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 29527.56, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "48 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Quadro RTX 8000", "accelerator_on-chip_memories": "", "accelerators_per_node": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56578502333763, "characteristics.scheduled_queries_per_second": 29527.56, "characteristics.scheduled_queries_per_second.normalized_per_core": 2952.7560000000003, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2952.7560000000003, "characteristics.word error rate": 7.434214976662383, "ck_system": "DSS8440_QuadroRTX8000x10_TRT", "ck_used": true, "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 11.0, cuDNN 8.0.2, cuBLAS 11.2.0, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "2.10GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.84 TB", "host_storage_type": "NVMe", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 10, "normalize_processors": 10, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/DSS8440_QuadroRTX8000x10_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": "docker 19.03.12, python 3.6.8, gcc 5.5.0, onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_QuadroRTX8000x10_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 29522, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "2f9b75a765e5ccd0", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 513837603, "90.00 percentile latency (ns)": 718449317, "95.00 percentile latency (ns)": 777142928, "97.00 percentile latency (ns)": 816656797, "99.00 percentile latency (ns)": 884274002, "99.90 percentile latency (ns)": 985587510, "Completed samples per second": 17247.19, "Max latency (ns)": 1136291795, "Mean latency (ns)": 524096061, "Min duration satisfied": "Yes", "Min latency (ns)": 48939618, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 17380.59, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 16, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 17380.59, "characteristics.scheduled_queries_per_second.normalized_per_core": 1086.286875, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1086.286875, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DSS8440_T4x16_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 11.0 Update 1", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.84 TB", "host_storage_type": "NVMe", "hw_notes": "ECC off", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 16, "normalize_processors": 16, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/DSS8440_T4x16_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_T4x16_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 17377, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "33247c2d24d7fa2d", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 479161312, "90.00 percentile latency (ns)": 682019011, "95.00 percentile latency (ns)": 739290641, "97.00 percentile latency (ns)": 775377631, "99.00 percentile latency (ns)": 836975441, "99.90 percentile latency (ns)": 923335850, "Completed samples per second": 4073.36, "Max latency (ns)": 996380562, "Mean latency (ns)": 495639342, "Min duration satisfied": "Yes", "Min latency (ns)": 74467444, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 4096.32, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.56127533879005, "characteristics.scheduled_queries_per_second": 4096.32, "characteristics.scheduled_queries_per_second.normalized_per_core": 1024.08, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1024.08, "characteristics.word error rate": 7.438724661209949, "ck_system": "R7515_T4x4_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 7.2.0.14, CUDA 11.0.207", "host_memory_capacity": "256 GB", "host_memory_configuration": "DDR-4", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "2.0GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "3.2 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/R7515_T4x4_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2.0.14, CUDA 11.0.27, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/R7515_T4x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 4100, "task": "speech recognition", "task2": "speech recognition", "total_cores": 120, "uid": "2c2b44da99e61d44", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 512395509, "90.00 percentile latency (ns)": 714867837, "95.00 percentile latency (ns)": 773878793, "97.00 percentile latency (ns)": 812103629, "99.00 percentile latency (ns)": 880801744, "99.90 percentile latency (ns)": 988461759, "Completed samples per second": 13016.52, "Max latency (ns)": 1145218416, "Mean latency (ns)": 523196362, "Min duration satisfied": "Yes", "Min latency (ns)": 52754019, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 13104.97, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 12, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.54549144287357, "characteristics.scheduled_queries_per_second": 13104.97, "characteristics.scheduled_queries_per_second.normalized_per_core": 1092.0808333333332, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1092.0808333333332, "characteristics.word error rate": 7.45450855712643, "ck_system": "DSS8440_T4x12_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 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC off", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 12, "normalize_processors": 12, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/DSS8440_T4x12_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/DSS8440_T4x12_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 13100, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "587e9d120bc4746b", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 515783389, "90.00 percentile latency (ns)": 647832892, "95.00 percentile latency (ns)": 695262342, "97.00 percentile latency (ns)": 720418345, "99.00 percentile latency (ns)": 758602555, "99.90 percentile latency (ns)": 833074573, "Completed samples per second": 11022.59, "Max latency (ns)": 871555052, "Mean latency (ns)": 514443779, "Min duration satisfied": "Yes", "Min latency (ns)": 77252301, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 11102.36, "accelerator_frequency": "1770MHz", "accelerator_host_interconnect": "PCI Express 3.0", "accelerator_interconnect": "PCI Express 3.0", "accelerator_interconnect_topology": "2 Accelerators per CPU", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Quadro RTX 6000", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55902049651627, "characteristics.scheduled_queries_per_second": 11102.36, "characteristics.scheduled_queries_per_second.normalized_per_core": 2775.59, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2775.59, "characteristics.word error rate": 7.440979503483732, "ck_system": "C4140_QuadroRTX6000x4_TRT", "ck_used": true, "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 11.0, cuDNN 8.0.2, cuBLAS 11.2.0, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "384 GB", "host_memory_configuration": "6x16GB DDR4-2666 HMA82GR7AFR8N-VK RDIMM ECC", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "1.25MB+20MB+27.5MB", "host_processor_core_count": 120, "host_processor_frequency": "2.40GHz", "host_processor_interconnect": "Ultra Path Interconnect", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1.6 TB (1x1.6TB Dell Express Flash PM1725a 1.6TB AIC)", "host_storage_type": "3D-TLC Solid State with PCIe NVME x8 Interface", "hw_notes": "ECC off. RTX6000 is available as a special config thru Dell DSS or OEM for PowerEdge C4140", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/C4140_QuadroRTX6000x4_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": "docker 19.03.12, python 3.6.8, gcc 5.5.0, onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/C4140_QuadroRTX6000x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 11100, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "c347878b6773e361", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 437038555, "90.00 percentile latency (ns)": 579047547, "95.00 percentile latency (ns)": 620101158, "97.00 percentile latency (ns)": 645534484, "99.00 percentile latency (ns)": 689675949, "99.90 percentile latency (ns)": 756231852, "Completed samples per second": 8961.58, "Max latency (ns)": 855708931, "Mean latency (ns)": 441790812, "Min duration satisfied": "Yes", "Min latency (ns)": 61499134, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 8995.57, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "48 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Quadro RTX 8000", "accelerator_on-chip_memories": "", "accelerators_per_node": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55225596969493, "characteristics.scheduled_queries_per_second": 8995.57, "characteristics.scheduled_queries_per_second.normalized_per_core": 2998.523333333333, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2998.523333333333, "characteristics.word error rate": 7.4477440303050795, "ck_system": "R7525_QuadroRTX8000x3_TRT", "ck_used": false, "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 11.0, cuDNN 8.0.2, cuBLAS 11.2.0, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 32, "host_processor_frequency": "2.50GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.84 TB", "host_storage_type": "NVMe", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/R7525_QuadroRTX8000x3_TRT", "number_of_nodes": 1, "operating_system": "CentOS Linux release 8.1.1911", "other_software_stack": "docker 19.03.12, python 3.6.8, gcc 5.5.0, onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_QuadroRTX8000x3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x Quadro RTX 8000)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 9000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "11b6d0e8174f41bb", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 432486632, "90.00 percentile latency (ns)": 649261441, "95.00 percentile latency (ns)": 722395277, "97.00 percentile latency (ns)": 761286650, "99.00 percentile latency (ns)": 866442166, "99.90 percentile latency (ns)": 1056202772, "Completed samples per second": 4172.32, "Max latency (ns)": 1291714115, "Mean latency (ns)": 452028862, "Min duration satisfied": "Yes", "Min latency (ns)": 65269453, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 4196.24, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.5680398656114, "characteristics.scheduled_queries_per_second": 4196.24, "characteristics.scheduled_queries_per_second.normalized_per_core": 1049.06, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1049.06, "characteristics.word error rate": 7.4319601343886, "ck_system": "R740_T4x4_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 7.2.0.14, CUDA 11.0.207", "host_memory_capacity": "384 GB", "host_memory_configuration": "DDR-4", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "3.0GHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "SSD", "hw_notes": "ECC on", "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)": 60000, "min_query_count": 270336, "mlperf_version": 0.7, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/DellEMC/results/R740_T4x4_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "TensorRT 7.2.0.14, CUDA 11.0.207, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 2513, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "N", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "DellEMC", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/DellEMC", "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/R740_T4x4_TRT", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 4200, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "e6c8798234002ab6", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "quantization, affine fusion" } ]