[ { "50.00 percentile latency (ns)": 348108579, "90.00 percentile latency (ns)": 483077324, "95.00 percentile latency (ns)": 522269465, "97.00 percentile latency (ns)": 545739655, "99.00 percentile latency (ns)": 592914992, "99.90 percentile latency (ns)": 683452311, "Completed samples per second": 42988.61, "Max latency (ns)": 854855039, "Mean latency (ns)": 356530902, "Min duration satisfied": "Yes", "Min latency (ns)": 36515769, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 43004.76, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.55225596969493, "characteristics.scheduled_queries_per_second": 43004.76, "characteristics.scheduled_queries_per_second.normalized_per_core": 10751.19, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10751.19, "characteristics.word error rate": 7.4477440303050795, "ck_system": "A100-PCIe-80GB_aarch64x4_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 80, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Ampere Altra Q80-30", "host_processors_per_node": 1, "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": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GB_aarch64x4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GB_aarch64x4_TRT", "system_name": "Gigabyte G242-P31 (4x A100-PCIe-80GB_aarch64, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 43000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 80, "uid": "8c5b62521853fbec", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 474963222, "90.00 percentile latency (ns)": 694854238, "95.00 percentile latency (ns)": 759368441, "97.00 percentile latency (ns)": 801151565, "99.00 percentile latency (ns)": 884274823, "99.90 percentile latency (ns)": 1030145645, "Completed samples per second": 36976.44, "Max latency (ns)": 1632175486, "Mean latency (ns)": 490325764, "Min duration satisfied": "Yes", "Min latency (ns)": 88700432, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 36999.32, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 36999.32, "characteristics.scheduled_queries_per_second.normalized_per_core": 4624.915, "characteristics.scheduled_queries_per_second.normalized_per_processor": 4624.915, "characteristics.word error rate": 7.4454891880312974, "ck_system": "A30x8_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": "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": "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": 1.1, "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 37000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "93435a91c945108c", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 316034589, "90.00 percentile latency (ns)": 447464566, "95.00 percentile latency (ns)": 485544404, "97.00 percentile latency (ns)": 509167287, "99.00 percentile latency (ns)": 556954279, "99.90 percentile latency (ns)": 641883611, "Completed samples per second": 89973.19, "Max latency (ns)": 827490959, "Mean latency (ns)": 324621496, "Min duration satisfied": "Yes", "Min latency (ns)": 35221705, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 90013.5, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIe-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": 90013.5, "characteristics.scheduled_queries_per_second.normalized_per_core": 11251.6875, "characteristics.scheduled_queries_per_second.normalized_per_processor": 11251.6875, "characteristics.word error rate": 7.4454891880312974, "ck_system": "A100-PCIe-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": "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": "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": 1.1, "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIe-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIe-80GBx8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 90000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "da2c31b4604a3e11", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 442472099, "90.00 percentile latency (ns)": 613253492, "95.00 percentile latency (ns)": 664488752, "97.00 percentile latency (ns)": 692539270, "99.00 percentile latency (ns)": 754233456, "99.90 percentile latency (ns)": 842861378, "Completed samples per second": 74966.28, "Max latency (ns)": 1027409685, "Mean latency (ns)": 450701861, "Min duration satisfied": "Yes", "Min latency (ns)": 65824135, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 75011.86, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.power": 2265.875666666665, "characteristics.power.normalized_per_core": 283.2344583333331, "characteristics.power.normalized_per_processor": 283.2344583333331, "characteristics.scheduled_queries_per_second": 75011.86, "characteristics.scheduled_queries_per_second.normalized_per_core": 9376.4825, "characteristics.scheduled_queries_per_second.normalized_per_processor": 9376.4825, "characteristics.word error rate": 7.4454891880312974, "ck_system": "A100-PCIex8_TRT_MaxQ", "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": "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": "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": 1.1, "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_MaxQ", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 75000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "3b10f02494286876", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 382531812, "90.00 percentile latency (ns)": 506631479, "95.00 percentile latency (ns)": 557950795, "97.00 percentile latency (ns)": 574584655, "99.00 percentile latency (ns)": 593806073, "99.90 percentile latency (ns)": 663378068, "Completed samples per second": 103968.71, "Max latency (ns)": 807582705, "Mean latency (ns)": 393497776, "Min duration satisfied": "Yes", "Min latency (ns)": 58047987, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 104011.92, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "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": 104011.92, "characteristics.scheduled_queries_per_second.normalized_per_core": 13001.49, "characteristics.scheduled_queries_per_second.normalized_per_processor": 13001.49, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DGX-A100_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": 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", "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 104000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "79c000d38e0c95af", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 434676280, "90.00 percentile latency (ns)": 583337194, "95.00 percentile latency (ns)": 622559010, "97.00 percentile latency (ns)": 654875288, "99.00 percentile latency (ns)": 705319173, "99.90 percentile latency (ns)": 784554786, "Completed samples per second": 44980.18, "Max latency (ns)": 934995563, "Mean latency (ns)": 438855226, "Min duration satisfied": "Yes", "Min latency (ns)": 86305280, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 45006.13, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5567656542425, "characteristics.scheduled_queries_per_second": 45006.13, "characteristics.scheduled_queries_per_second.normalized_per_core": 11251.5325, "characteristics.scheduled_queries_per_second.normalized_per_processor": 11251.5325, "characteristics.word error rate": 7.4432343457575145, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_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": "512 GB", "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": 1, "host_storage_capacity": "10 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": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 45000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "89dd38546815d07e", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 496141731, "90.00 percentile latency (ns)": 725313303, "95.00 percentile latency (ns)": 802707798, "97.00 percentile latency (ns)": 842667114, "99.00 percentile latency (ns)": 941622322, "99.90 percentile latency (ns)": 1077118779, "Completed samples per second": 949.21, "Max latency (ns)": 1147154557, "Mean latency (ns)": 511273276, "Min duration satisfied": "Yes", "Min latency (ns)": 98485294, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 949.61, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30 (1x1g.6gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.54774628514735, "characteristics.scheduled_queries_per_second": 949.61, "characteristics.scheduled_queries_per_second.normalized_per_core": 949.61, "characteristics.scheduled_queries_per_second.normalized_per_processor": 949.61, "characteristics.word error rate": 7.452253714852645, "ck_system": "A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "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": "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": "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": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30-MIG_1x1g.6gb_TRT_HeteroMultiUse", "system_name": "Gigabyte G482-Z54 (1x A30-MIG-1x1g.6gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 950, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "72ab84c0a1ada581", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 419433691, "90.00 percentile latency (ns)": 648524119, "95.00 percentile latency (ns)": 715071584, "97.00 percentile latency (ns)": 753768904, "99.00 percentile latency (ns)": 805731202, "99.90 percentile latency (ns)": 843052967, "Completed samples per second": 1299.01, "Max latency (ns)": 910583424, "Mean latency (ns)": 447884229, "Min duration satisfied": "Yes", "Min latency (ns)": 82002477, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 1299.7, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (1x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.54549144287357, "characteristics.scheduled_queries_per_second": 1299.7, "characteristics.scheduled_queries_per_second.normalized_per_core": 1299.7, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1299.7, "characteristics.word error rate": 7.45450855712643, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "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": 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", "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": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "system_name": "NVIDIA DGX A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 1300, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "958f36fe6b580554", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 440233634, "90.00 percentile latency (ns)": 595426466, "95.00 percentile latency (ns)": 636037350, "97.00 percentile latency (ns)": 671227229, "99.00 percentile latency (ns)": 725879939, "99.90 percentile latency (ns)": 811015049, "Completed samples per second": 43388.84, "Max latency (ns)": 936382869, "Mean latency (ns)": 443135303, "Min duration satisfied": "Yes", "Min latency (ns)": 87693388, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 43406.14, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5567656542425, "characteristics.power": 1293.7970000000003, "characteristics.power.normalized_per_core": 323.44925000000006, "characteristics.power.normalized_per_processor": 323.44925000000006, "characteristics.scheduled_queries_per_second": 43406.14, "characteristics.scheduled_queries_per_second.normalized_per_core": 10851.535, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10851.535, "characteristics.word error rate": 7.4432343457575145, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "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": "512 GB", "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": 1, "host_storage_capacity": "10 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": 1.1, "network_speed_mbit": "", "nics_enabled_connected": "", "nics_enabled_firmware": "", "nics_enabled_os": "", "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 43400, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "9f11b46db4fa8bae", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 444767561, "90.00 percentile latency (ns)": 593595088, "95.00 percentile latency (ns)": 635041559, "97.00 percentile latency (ns)": 665138529, "99.00 percentile latency (ns)": 701886132, "99.90 percentile latency (ns)": 783093764, "Completed samples per second": 85964.27, "Max latency (ns)": 945931587, "Mean latency (ns)": 454118229, "Min duration satisfied": "Yes", "Min latency (ns)": 54612119, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 86014.56, "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": 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": 86014.56, "characteristics.scheduled_queries_per_second.normalized_per_core": 10751.82, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10751.82, "characteristics.word error rate": 7.4454891880312974, "ck_system": "A100-PCIex8_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": "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": "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": 1.1, "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 86000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "45226423c91839fc", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 411846726, "90.00 percentile latency (ns)": 575046273, "95.00 percentile latency (ns)": 613818277, "97.00 percentile latency (ns)": 650725351, "99.00 percentile latency (ns)": 694901760, "99.90 percentile latency (ns)": 776055952, "Completed samples per second": 87967.14, "Max latency (ns)": 896151456, "Mean latency (ns)": 422317016, "Min duration satisfied": "Yes", "Min latency (ns)": 45781428, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 88013.5, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "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": 92.55225596969493, "characteristics.power": 3457.1848333333305, "characteristics.power.normalized_per_core": 432.1481041666663, "characteristics.power.normalized_per_processor": 432.1481041666663, "characteristics.scheduled_queries_per_second": 88013.5, "characteristics.scheduled_queries_per_second.normalized_per_core": 11001.6875, "characteristics.scheduled_queries_per_second.normalized_per_processor": 11001.6875, "characteristics.word error rate": 7.4477440303050795, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "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": 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", "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "system_name": "NVIDIA DGX A100 (8x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 88000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "304758947bb4ea90", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 339007358, "90.00 percentile latency (ns)": 506631258, "95.00 percentile latency (ns)": 547556617, "97.00 percentile latency (ns)": 568939497, "99.00 percentile latency (ns)": 619323961, "99.90 percentile latency (ns)": 674734321, "Completed samples per second": 22592.03, "Max latency (ns)": 776211910, "Mean latency (ns)": 355246603, "Min duration satisfied": "Yes", "Min latency (ns)": 41153916, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 22599.48, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "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": 92.54774628514735, "characteristics.scheduled_queries_per_second": 22599.48, "characteristics.scheduled_queries_per_second.normalized_per_core": 2824.935, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2824.935, "characteristics.word error rate": 7.452253714852645, "ck_system": "A10x8_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": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "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": 1.1, "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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/NVIDIA/results/A10x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 20.04.4", "other_hardware": "", "other_software_stack": "TensorRT 8.0.1, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 22605, "task": "speech recognition", "task2": "speech recognition", "total_cores": 56, "uid": "b45399b8f34ee97d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 404221492, "90.00 percentile latency (ns)": 506837749, "95.00 percentile latency (ns)": 541523496, "97.00 percentile latency (ns)": 557760464, "99.00 percentile latency (ns)": 596795765, "99.90 percentile latency (ns)": 687295592, "Completed samples per second": 12747.6, "Max latency (ns)": 826785495, "Mean latency (ns)": 408674330, "Min duration satisfied": "Yes", "Min latency (ns)": 69935076, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 12751.37, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 12751.37, "characteristics.scheduled_queries_per_second.normalized_per_core": 12751.37, "characteristics.scheduled_queries_per_second.normalized_per_processor": 12751.37, "characteristics.word error rate": 7.449998872578862, "ck_system": "DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "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": "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", "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": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "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/DGX-A100_A100-SXM-80GBx1_TRT_datacenter", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 12750, "task": "speech recognition", "task2": "speech recognition", "total_cores": 240, "uid": "797600ac5e34badf", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 435815083, "90.00 percentile latency (ns)": 711795301, "95.00 percentile latency (ns)": 803243774, "97.00 percentile latency (ns)": 863719866, "99.00 percentile latency (ns)": 984784065, "99.90 percentile latency (ns)": 1213087125, "Completed samples per second": 36959.37, "Max latency (ns)": 1667955553, "Mean latency (ns)": 463365114, "Min duration satisfied": "Yes", "Min latency (ns)": 79166385, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 36990.65, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 36990.65, "characteristics.scheduled_queries_per_second.normalized_per_core": 4623.83125, "characteristics.scheduled_queries_per_second.normalized_per_processor": 4623.83125, "characteristics.word error rate": 7.4454891880312974, "ck_system": "A30x8_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 7.2.3, CUDA 11.1", "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": "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": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.46, DALI 0.30.0", "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": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A30x8_TRT", "system_name": "Gigabyte G482-Z54 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 37000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "f948958427a5f7ba", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 465018026, "90.00 percentile latency (ns)": 631518474, "95.00 percentile latency (ns)": 674436306, "97.00 percentile latency (ns)": 704726130, "99.00 percentile latency (ns)": 762301641, "99.90 percentile latency (ns)": 857393613, "Completed samples per second": 74929.9, "Max latency (ns)": 1070568765, "Mean latency (ns)": 473181294, "Min duration satisfied": "Yes", "Min latency (ns)": 68123557, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 74974.26, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.55000112742113, "characteristics.power": 2324.844833333333, "characteristics.power.normalized_per_core": 290.60560416666664, "characteristics.power.normalized_per_processor": 290.60560416666664, "characteristics.scheduled_queries_per_second": 74974.26, "characteristics.scheduled_queries_per_second.normalized_per_core": 9371.7825, "characteristics.scheduled_queries_per_second.normalized_per_processor": 9371.7825, "characteristics.word error rate": 7.449998872578862, "ck_system": "A100-PCIex8_TRT_MaxQ", "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 7.2.3, CUDA 11.1", "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": "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": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT_MaxQ", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 75000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "5d7ba66f880d463d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 340479745, "90.00 percentile latency (ns)": 466703116, "95.00 percentile latency (ns)": 494062041, "97.00 percentile latency (ns)": 527295865, "99.00 percentile latency (ns)": 558058229, "99.90 percentile latency (ns)": 637633708, "Completed samples per second": 103940.0, "Max latency (ns)": 887243027, "Mean latency (ns)": 351979660, "Min duration satisfied": "Yes", "Min latency (ns)": 85687884, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 103985.94, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-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": 103985.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 12998.2425, "characteristics.scheduled_queries_per_second.normalized_per_processor": 12998.2425, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DGX-A100_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 7.2.3, CUDA 11.1", "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": "15 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": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 104000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "8bc1d06cb2993c87", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 402175070, "90.00 percentile latency (ns)": 557138200, "95.00 percentile latency (ns)": 602328857, "97.00 percentile latency (ns)": 633065015, "99.00 percentile latency (ns)": 686676258, "99.90 percentile latency (ns)": 769930273, "Completed samples per second": 43367.11, "Max latency (ns)": 918308679, "Mean latency (ns)": 410280991, "Min duration satisfied": "Yes", "Min latency (ns)": 74742905, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 43388.92, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 43388.92, "characteristics.scheduled_queries_per_second.normalized_per_core": 10847.23, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10847.23, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_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 7.2.3, CUDA 11.1", "host_memory_capacity": "512 GB", "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": 1, "host_storage_capacity": "10 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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 43400, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "2e292cdf1a4285a3", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 428919636, "90.00 percentile latency (ns)": 656455387, "95.00 percentile latency (ns)": 722658304, "97.00 percentile latency (ns)": 760696817, "99.00 percentile latency (ns)": 812670389, "99.90 percentile latency (ns)": 851136895, "Completed samples per second": 1302.0, "Max latency (ns)": 911529149, "Mean latency (ns)": 456915724, "Min duration satisfied": "Yes", "Min latency (ns)": 113907082, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 1302.5, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB (1x1g.10gb MIG)", "accelerator_on-chip_memories": "", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.54549144287357, "characteristics.scheduled_queries_per_second": 1302.5, "characteristics.scheduled_queries_per_second.normalized_per_core": 1302.5, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1302.5, "characteristics.word error rate": 7.45450855712643, "ck_system": "DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "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 7.2.3, CUDA 11.1", "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": "15 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": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GB-MIG_1x1g.10gb_TRT_HeteroMultiUse", "system_name": "NVIDIA DGX-A100 (1x A100-SXM-80GB-MIG-1x1g.10gb, TensorRT, HeteroMultiUse)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 1300, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "6f42e1961826fc7b", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 456417804, "90.00 percentile latency (ns)": 608819784, "95.00 percentile latency (ns)": 650319738, "97.00 percentile latency (ns)": 681596193, "99.00 percentile latency (ns)": 729929565, "99.90 percentile latency (ns)": 811813590, "Completed samples per second": 43363.35, "Max latency (ns)": 1023475649, "Mean latency (ns)": 464202866, "Min duration satisfied": "Yes", "Min latency (ns)": 100527186, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 43388.92, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.power": 1313.5377703826957, "characteristics.power.normalized_per_core": 328.38444259567393, "characteristics.power.normalized_per_processor": 328.38444259567393, "characteristics.scheduled_queries_per_second": 43388.92, "characteristics.scheduled_queries_per_second.normalized_per_core": 10847.23, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10847.23, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "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 7.2.3, CUDA 11.1", "host_memory_capacity": "512 GB", "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": 1, "host_storage_capacity": "10 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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-Station-A100_A100-SXM-80GBx4_TRT_MaxQ", "system_name": "NVIDIA DGX Station A100 (4x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 43400, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "bdd13074724aeec2", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 429707784, "90.00 percentile latency (ns)": 576847831, "95.00 percentile latency (ns)": 616457903, "97.00 percentile latency (ns)": 648642417, "99.00 percentile latency (ns)": 688787698, "99.90 percentile latency (ns)": 774857086, "Completed samples per second": 85927.67, "Max latency (ns)": 987563779, "Mean latency (ns)": 438909367, "Min duration satisfied": "Yes", "Min latency (ns)": 95045142, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 85985.32, "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": 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": 85985.32, "characteristics.scheduled_queries_per_second.normalized_per_core": 10748.165, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10748.165, "characteristics.word error rate": 7.449998872578862, "ck_system": "A100-PCIex8_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 7.2.3, CUDA 11.1", "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": "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": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A100-PCIex8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex8_TRT", "system_name": "Gigabyte G482-Z54 (8x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 86000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "4bf887280891cecb", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 356208283, "90.00 percentile latency (ns)": 493535012, "95.00 percentile latency (ns)": 530823310, "97.00 percentile latency (ns)": 550249191, "99.00 percentile latency (ns)": 606445581, "99.90 percentile latency (ns)": 684686288, "Completed samples per second": 87956.3, "Max latency (ns)": 839687239, "Mean latency (ns)": 362620595, "Min duration satisfied": "Yes", "Min latency (ns)": 78947483, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 87983.67, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "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.power": 3550.9643926788626, "characteristics.power.normalized_per_core": 443.8705490848578, "characteristics.power.normalized_per_processor": 443.8705490848578, "characteristics.scheduled_queries_per_second": 87983.67, "characteristics.scheduled_queries_per_second.normalized_per_core": 10997.95875, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10997.95875, "characteristics.word error rate": 7.449998872578862, "ck_system": "DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "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 7.2.3, CUDA 11.1", "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": "15 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": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM-80GBx8_TRT_MaxQ", "system_name": "NVIDIA DGX-A100 (8x A100-SXM-80GB, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 88000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "4ebc425460faca7b", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 367665118, "90.00 percentile latency (ns)": 582915077, "95.00 percentile latency (ns)": 663513168, "97.00 percentile latency (ns)": 725437164, "99.00 percentile latency (ns)": 902635512, "99.90 percentile latency (ns)": 1336772331, "Completed samples per second": 22589.89, "Max latency (ns)": 1891007676, "Mean latency (ns)": 391739217, "Min duration satisfied": "Yes", "Min latency (ns)": 39291888, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 22596.98, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "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": 22596.98, "characteristics.scheduled_queries_per_second.normalized_per_core": 2824.6225, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2824.6225, "characteristics.word error rate": 7.449998872578862, "ck_system": "A10x8_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 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": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "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": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/NVIDIA/results/A10x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "Ubuntu 18.04.4", "other_hardware": "", "other_software_stack": "TensorRT 7.2.3, CUDA 11.1, cuDNN 8.1.1, Driver 460.32.03, DALI 0.30.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A10x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 22605, "task": "speech recognition", "task2": "speech recognition", "total_cores": 56, "uid": "ab2ae39e4d54908d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 387032903, "90.00 percentile latency (ns)": 580194585, "95.00 percentile latency (ns)": 652862546, "97.00 percentile latency (ns)": 717233102, "99.00 percentile latency (ns)": 858575434, "99.90 percentile latency (ns)": 1052574750, "Completed samples per second": 19913.29, "Max latency (ns)": 1643710226, "Mean latency (ns)": 412939737, "Min duration satisfied": "Yes", "Min latency (ns)": 63260015, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 20001.64, "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": 20, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 20001.64, "characteristics.scheduled_queries_per_second.normalized_per_core": 1000.082, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1000.082, "characteristics.word error rate": 7.4454891880312974, "ck_system": "T4x20_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 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 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": 20, "normalize_processors": 20, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x20_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x20_TRT", "system_name": "Supermicro 6049GP-TRT-OTO-29 (20x T4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 20000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 56, "uid": "763a80ef40041f19", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 588143669, "90.00 percentile latency (ns)": 797573042, "95.00 percentile latency (ns)": 859560431, "97.00 percentile latency (ns)": 897061114, "99.00 percentile latency (ns)": 965351003, "99.90 percentile latency (ns)": 1075308877, "Completed samples per second": 14908.79, "Max latency (ns)": 1274097795, "Mean latency (ns)": 601011593, "Min duration satisfied": "Yes", "Min latency (ns)": 57109498, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 15002.23, "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": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 15002.23, "characteristics.scheduled_queries_per_second.normalized_per_core": 7501.115, "characteristics.scheduled_queries_per_second.normalized_per_processor": 7501.115, "characteristics.word error rate": 7.449998872578862, "ck_system": "A100-PCIex2_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 11.0 Update 1", "host_memory_capacity": "768 GB", "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": "4 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)": 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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/A100-PCIex2_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/A100-PCIex2_TRT", "system_name": "Gigabyte G482-Z52 (2x A100-PCIe, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 15000, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "dc1e984df2ba1575", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 423459990, "90.00 percentile latency (ns)": 591373621, "95.00 percentile latency (ns)": 632371322, "97.00 percentile latency (ns)": 659565582, "99.00 percentile latency (ns)": 725931703, "99.90 percentile latency (ns)": 829095464, "Completed samples per second": 71178.43, "Max latency (ns)": 1019328780, "Mean latency (ns)": 436704634, "Min duration satisfied": "Yes", "Min latency (ns)": 56456456, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 71510.03, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-SXM4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.54774628514735, "characteristics.scheduled_queries_per_second": 71510.03, "characteristics.scheduled_queries_per_second.normalized_per_core": 8938.75375, "characteristics.scheduled_queries_per_second.normalized_per_processor": 8938.75375, "characteristics.word error rate": 7.452253714852645, "ck_system": "DGX-A100_A100-SXM4x8_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 11.0 Update 1", "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)": 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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/DGX-A100_A100-SXM4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DGX-A100_A100-SXM4x8_TRT", "system_name": "NVIDIA DGX-A100 (8x A100-SXM4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 71500, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "4d129209ac9eac03", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 504269488, "90.00 percentile latency (ns)": 771246527, "95.00 percentile latency (ns)": 845126237, "97.00 percentile latency (ns)": 905690774, "99.00 percentile latency (ns)": 999063308, "99.90 percentile latency (ns)": 1165655685, "Completed samples per second": 8595.03, "Max latency (ns)": 1356784903, "Mean latency (ns)": 534611242, "Min duration satisfied": "Yes", "Min latency (ns)": 44280796, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 8659.56, "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": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 92.55225596969493, "characteristics.scheduled_queries_per_second": 8659.56, "characteristics.scheduled_queries_per_second.normalized_per_core": 1082.445, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1082.445, "characteristics.word error rate": 7.4477440303050795, "ck_system": "T4x8_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 11.0 Update 1", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 28, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 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/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/NVIDIA/results/T4x8_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.4", "other_software_stack": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 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": "NVIDIA", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/NVIDIA", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/T4x8_TRT", "system_name": "Supermicro 4029GP-TRT-OTO-28 (8x T4, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 8660, "task": "speech recognition", "task2": "speech recognition", "total_cores": 56, "uid": "95f8a094b2d1e1ae", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" } ]