[ { "50.00 percentile latency (ns)": 443127891, "90.00 percentile latency (ns)": 607817880, "95.00 percentile latency (ns)": 630240160, "97.00 percentile latency (ns)": 688589289, "99.00 percentile latency (ns)": 707406774, "99.90 percentile latency (ns)": 790473453, "Completed samples per second": 32556.39, "Max latency (ns)": 898314000, "Mean latency (ns)": 468202295, "Min duration satisfied": "Yes", "Min latency (ns)": 57535297, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 32577.05, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "40 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A100-PCIE-40GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.54549144287357, "characteristics.scheduled_queries_per_second": 32577.05, "characteristics.scheduled_queries_per_second.normalized_per_core": 10859.016666666666, "characteristics.scheduled_queries_per_second.normalized_per_processor": 10859.016666666666, "characteristics.word error rate": 7.45450855712643, "ck_system": "R7525_A100-PCIE-40GBx3_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": "512 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 32, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7502", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "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": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R7525_A100-PCIE-40GBx3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A100-PCIE-40GBx3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A100-PCIE-40GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 32580, "task": "speech recognition", "task2": "speech recognition", "total_cores": 64, "uid": "441885c019cf0e1d", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 398023774, "90.00 percentile latency (ns)": 540956588, "95.00 percentile latency (ns)": 587238260, "97.00 percentile latency (ns)": 612792945, "99.00 percentile latency (ns)": 629150271, "99.90 percentile latency (ns)": 704284571, "Completed samples per second": 48483.57, "Max latency (ns)": 803727106, "Mean latency (ns)": 420617026, "Min duration satisfied": "Yes", "Min latency (ns)": 39980330, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 48504.44, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 48504.44, "characteristics.scheduled_queries_per_second.normalized_per_core": 12126.11, "characteristics.scheduled_queries_per_second.normalized_per_processor": 12126.11, "characteristics.word error rate": 7.449998872578862, "ck_system": "R750xa_A100-PCIE-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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6338 CPU @ 2.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3.5 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R750xa_A100-PCIE-80GBx4_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R750xa_A100-PCIE-80GBx4_TRT", "system_name": "Dell EMC PowerEdge R750xa (4x A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 48500, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "181459829e47b497", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 153915871, "90.00 percentile latency (ns)": 216466463, "95.00 percentile latency (ns)": 252229324, "97.00 percentile latency (ns)": 259722743, "99.00 percentile latency (ns)": 267813246, "99.90 percentile latency (ns)": 281790720, "Completed samples per second": 3301.67, "Max latency (ns)": 334832425, "Mean latency (ns)": 158293487, "Min duration satisfied": "Yes", "Min latency (ns)": 50453863, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 3302.19, "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.accuracy": 92.54549144287357, "characteristics.power": 540.312066666667, "characteristics.power.normalized_per_core": 270.1560333333335, "characteristics.power.normalized_per_processor": 270.1560333333335, "characteristics.scheduled_queries_per_second": 3302.19, "characteristics.scheduled_queries_per_second.normalized_per_core": 1651.095, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1651.095, "characteristics.word error rate": 7.45450855712643, "ck_system": "XE2420_A10x2_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, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "air", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6252 CPU @ 2.10GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "5.00.00.00", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE2420_A10x2_TRT_MaxQ", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_settings": "closed/Dell/power/XE2420_A10x2_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_A10x2_TRT_MaxQ", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, MaxQ, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 3300, "task": "speech recognition", "task2": "speech recognition", "total_cores": 48, "uid": "0a2cc03c6d4139b0", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 537416303, "90.00 percentile latency (ns)": 742691063, "95.00 percentile latency (ns)": 790249792, "97.00 percentile latency (ns)": 826606213, "99.00 percentile latency (ns)": 880105485, "99.90 percentile latency (ns)": 965841808, "Completed samples per second": 117746.1, "Max latency (ns)": 1172543151, "Mean latency (ns)": 548747507, "Min duration satisfied": "Yes", "Min latency (ns)": 45704365, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 117865.66, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80 GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-PCIE-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 10, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.5545108119687, "characteristics.scheduled_queries_per_second": 117865.66, "characteristics.scheduled_queries_per_second.normalized_per_core": 11786.566, "characteristics.scheduled_queries_per_second.normalized_per_processor": 11786.566, "characteristics.word error rate": 7.4454891880312974, "ck_system": "DSS8440_A100-PCIE-80GBx10_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": " 768GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "1.8 TB", "host_storage_type": "SSD", "hw_notes": "", "informal_model": "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": 10, "normalize_processors": 10, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/DSS8440_A100-PCIE-80GBx10_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A100-PCIE-80GBx10_TRT", "system_name": "Dell EMC DSS 8440 (10x NVIDIA A100-PCIE-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 117851, "task": "speech recognition", "task2": "speech recognition", "total_cores": 48, "uid": "cc8bef9498e31b52", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 466073592, "90.00 percentile latency (ns)": 727422219, "95.00 percentile latency (ns)": 810936575, "97.00 percentile latency (ns)": 872434414, "99.00 percentile latency (ns)": 991519818, "99.90 percentile latency (ns)": 1265565419, "Completed samples per second": 35284.26, "Max latency (ns)": 1742274157, "Mean latency (ns)": 490005325, "Min duration satisfied": "Yes", "Min latency (ns)": 58953113, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 35299.01, "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": 92.55000112742113, "characteristics.scheduled_queries_per_second": 35299.01, "characteristics.scheduled_queries_per_second.normalized_per_core": 4412.37625, "characteristics.scheduled_queries_per_second.normalized_per_processor": 4412.37625, "characteristics.word error rate": 7.449998872578862, "ck_system": "DSS8440_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 8.0.2, CUDA 11.3", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 48, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe", "hw_notes": "", "informal_model": "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/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/DSS8440_A30x8_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.2", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/DSS8440_A30x8_TRT", "system_name": "Dell EMC DSS 8440 (8x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 35300, "task": "speech recognition", "task2": "speech recognition", "total_cores": 96, "uid": "eb4a43a9d4869064", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 351581473, "90.00 percentile latency (ns)": 509820992, "95.00 percentile latency (ns)": 557795370, "97.00 percentile latency (ns)": 586769610, "99.00 percentile latency (ns)": 644434079, "99.90 percentile latency (ns)": 739150474, "Completed samples per second": 10887.37, "Max latency (ns)": 899944851, "Mean latency (ns)": 362262747, "Min duration satisfied": "Yes", "Min latency (ns)": 82373007, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 10891.59, "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "HBM2", "accelerator_model_name": "NVIDIA A30", "accelerator_on-chip_memories": "", "accelerators_per_node": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "", "characteristics.accuracy": 92.55000112742113, "characteristics.scheduled_queries_per_second": 10891.59, "characteristics.scheduled_queries_per_second.normalized_per_core": 3630.53, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3630.53, "characteristics.word error rate": 7.449998872578862, "ck_system": "R7525_A30x3_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": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7763", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "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": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/R7525_A30x3_TRT", "number_of_nodes": 1, "number_of_type_nics_installed": "", "operating_system": "CentOS 8.4.2105", "other_hardware": "", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.42.01, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_supply_details": "", "power_supply_quantity_and_rating_watts": "", "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/R7525_A30x3_TRT", "system_name": "Dell EMC PowerEdge R7525 (3x A30, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 10890, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "e0acd593682a43f1", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 463738474, "90.00 percentile latency (ns)": 655749992, "95.00 percentile latency (ns)": 716523594, "97.00 percentile latency (ns)": 751083842, "99.00 percentile latency (ns)": 794691169, "99.90 percentile latency (ns)": 882905550, "Completed samples per second": 6899.69, "Max latency (ns)": 1026928934, "Mean latency (ns)": 471241208, "Min duration satisfied": "Yes", "Min latency (ns)": 43378328, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 6902.14, "accelerator_cooling_type": "passive", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA A10", "accelerator_on-chip_memories": "", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.12.3", "characteristics.accuracy": 92.5567656542425, "characteristics.scheduled_queries_per_second": 6902.14, "characteristics.scheduled_queries_per_second.normalized_per_core": 3451.07, "characteristics.scheduled_queries_per_second.normalized_per_processor": 3451.07, "characteristics.word error rate": 7.4432343457575145, "ck_system": "XE2420_A10x2_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", "filesystem": "ext3/ext4", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "air", "host_memory_capacity": "384 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Gold 6252 CPU @ 2.10GHz", "host_processors_per_node": 2, "host_storage_capacity": "4 TB", "host_storage_type": "NVMe SSD", "hw_notes": "ECC on", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "5.00.00.00", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE2420_A10x2_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_settings": "closed/Dell/power/XE2420_A10x2_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "2x2000W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE2420_A10x2_TRT", "system_name": "Dell EMC PowerEdge XE2420 (2x A10, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 6900, "task": "speech recognition", "task2": "speech recognition", "total_cores": 48, "uid": "4861adfd0e8285ac", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 422344057, "90.00 percentile latency (ns)": 529303234, "95.00 percentile latency (ns)": 566506657, "97.00 percentile latency (ns)": 582514961, "99.00 percentile latency (ns)": 623719903, "99.90 percentile latency (ns)": 682243563, "Completed samples per second": 55172.96, "Max latency (ns)": 792076350, "Mean latency (ns)": 423713592, "Min duration satisfied": "Yes", "Min latency (ns)": 36806111, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "RNNT SERVER", "Scenario": "server", "Scheduled samples per second": 55204.1, "accelerator_cooling_type": "", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "80GB", "accelerator_memory_configuration": "HBM2e", "accelerator_model_name": "NVIDIA A100-SXM-80GB", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "boot_firmware_version": "2.2.5", "characteristics.accuracy": 92.55000112742113, "characteristics.power": 2629.9176372712145, "characteristics.power.normalized_per_core": 657.4794093178036, "characteristics.power.normalized_per_processor": 657.4794093178036, "characteristics.scheduled_queries_per_second": 55204.1, "characteristics.scheduled_queries_per_second.normalized_per_core": 13801.025, "characteristics.scheduled_queries_per_second.normalized_per_processor": 13801.025, "characteristics.word error rate": 7.449998872578862, "ck_system": "XE8545_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, "division": "closed", "filesystem": "ext3/ext4", "formal_model": "rnn-t", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "TensorRT 8.0.2, CUDA 11.3", "host_cooling_type": "", "host_memory_capacity": "1 TB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 64, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7763", "host_processors_per_node": 2, "host_storage_capacity": "3 TB", "host_storage_type": "NVMe SSD", "hw_notes": "500W A100-SXM-80GB", "informal_model": "rnnt", "input_data_types": "fp16", "key.accuracy": "characteristics.accuracy", "management_firmware_version": "4.40.40.151", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "mlperf_version": 1.1, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Dell/results/XE8545_A100-SXM-80GBx4_TRT", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.2", "other_software_stack": "TensorRT 8.0.2, CUDA 11.3, cuDNN 8.2.1, Driver 470.57.02, DALI 0.31.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 2513, "power_management": "", "power_settings": "closed/Dell/power/XE8545_A100-SXM-80GBx4_power_settings.md", "print_timestamps": 0, "problem": false, "psu_details": "4x2400W", "qsl_rng_seed": 1624344308455410291, "retraining": "N", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt", "status": "available", "submitter": "Dell", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Dell", "sw_notes": "", "system_cooling_type": "air", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/XE8545_A100-SXM-80GBx4_TRT", "system_name": "Dell EMC PowerEdge XE8545 (4x A100-SXM-80GB, TensorRT)", "system_type": "datacenter", "target_latency (ns)": 1000000000, "target_qps": 55200, "task": "speech recognition", "task2": "speech recognition", "total_cores": 128, "uid": "46904a5ba7e9b193", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "quantization, affine fusion" } ]