[ { "50.00 percentile latency (ns)": 3808756, "90.00 percentile latency (ns)": 3844327, "90th percentile latency (ns)": 3844327, "95.00 percentile latency (ns)": 3854780, "97.00 percentile latency (ns)": 3862355, "99.00 percentile latency (ns)": 3879092, "99.90 percentile latency (ns)": 3903938, "Max latency (ns)": 8142947, "Mean latency (ns)": 3799688, "Min duration satisfied": "Yes", "Min latency (ns)": 3588748, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 258.94, "QPS w/o loadgen overhead": 263.18, "Result is": "VALID", "SUT name": "LWIS_Server", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "24 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA TITAN RTX", "accelerator_on-chip_memories": "", "accelerators_per_node": 4, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.90th_percentile_latency_ms": 3.844327, "characteristics.90th_percentile_latency_ns": 3844327.0, "characteristics.90th_percentile_latency_s": 0.003844327, "characteristics.90th_percentile_latency_us": 3844.327, "characteristics.mAP": 20.067, "ck_system": "TitanRTXx4", "ck_used": false, "cooling": "watercooled", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.90th_percentile_latency_ms", "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 24, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) 8268", "host_processors_per_node": 2, "host_storage_capacity": "3.84 TB", "host_storage_type": "NVMe SSD", "hw_notes": "", "informal_model": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1024, "mlperf_version": 0.5, "normalize_cores": 4, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/TitanRTXx4", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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/TitanRTXx4", "system_name": "SCAN 3XS DBP T496X2 Fluid", "target_latency (ns)": 0, "target_qps": 236.032, "task": "object detection", "task2": "object detection", "total_cores": 48, "uid": "1e9a2a32b4c448bd", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 1486539, "90.00 percentile latency (ns)": 1500587, "90th percentile latency (ns)": 1500587, "95.00 percentile latency (ns)": 1508044, "97.00 percentile latency (ns)": 1514284, "99.00 percentile latency (ns)": 1527501, "99.90 percentile latency (ns)": 1556623, "Max latency (ns)": 2144460, "Mean latency (ns)": 1488133, "Min duration satisfied": "Yes", "Min latency (ns)": 1450825, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 664.36, "QPS w/o loadgen overhead": 671.98, "Result is": "VALID", "SUT name": "LWIS_Server", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "Shared with host", "accelerator_memory_configuration": "SRAM", "accelerator_model_name": "NVIDIA Xavier", "accelerator_on-chip_memories": "1MB (128KB/SM) L1 + 512KB L2 + 4MB (DLA)", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.90th_percentile_latency_ms": 1.500587, "characteristics.90th_percentile_latency_ns": 1500587.0, "characteristics.90th_percentile_latency_s": 0.001500587, "characteristics.90th_percentile_latency_us": 1500.587, "characteristics.mAP": 22.9, "ck_system": "Xavier", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.90th_percentile_latency_ms", "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "JetPack 4.3 DP, TensorRT 6.0, cuDNN 7.6.3, CUDA 10.0, cub 1.8.0", "host_memory_capacity": "16 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "8MB L2 (2MB per dual cluster)/4 MB L3 (shared)", "host_processor_core_count": 8, "host_processor_frequency": "2265.5 MHz", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32 GB", "host_storage_type": "eMMC 5.1", "hw_notes": "GPU and both DLAs are used in Offline and MultiStream scenarios", "informal_model": "ssd-small", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1024, "mlperf_version": 0.5, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/Xavier", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "pycuda 2019.1, pytorch 1.1, torchvision 0.2.2.post3", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 256, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "frozen_inference_graph.pb", "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/Xavier", "system_name": "NVIDIA Jetson AGX Xavier", "target_latency (ns)": 0, "target_qps": 616.903, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "9f72b5d4c51a69d8", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 29350340, "90.00 percentile latency (ns)": 29428616, "90th percentile latency (ns)": 29428616, "95.00 percentile latency (ns)": 29466122, "97.00 percentile latency (ns)": 29493579, "99.00 percentile latency (ns)": 29695350, "99.90 percentile latency (ns)": 29821852, "Max latency (ns)": 30092041, "Mean latency (ns)": 29360819, "Min duration satisfied": "Yes", "Min latency (ns)": 29238558, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 34.04, "QPS w/o loadgen overhead": 34.06, "Result is": "VALID", "SUT name": "LWIS_Server", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "Shared with host", "accelerator_memory_configuration": "SRAM", "accelerator_model_name": "NVIDIA Xavier", "accelerator_on-chip_memories": "1MB (128KB/SM) L1 + 512KB L2 + 4MB (DLA)", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.90th_percentile_latency_ms": 29.428616, "characteristics.90th_percentile_latency_ns": 29428616.0, "characteristics.90th_percentile_latency_s": 0.029428616, "characteristics.90th_percentile_latency_us": 29428.616, "characteristics.mAP": 20.067, "ck_system": "Xavier", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.90th_percentile_latency_ms", "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "JetPack 4.3 DP, TensorRT 6.0, cuDNN 7.6.3, CUDA 10.0, cub 1.8.0", "host_memory_capacity": "16 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "8MB L2 (2MB per dual cluster)/4 MB L3 (shared)", "host_processor_core_count": 8, "host_processor_frequency": "2265.5 MHz", "host_processor_interconnect": "", "host_processor_model_name": "NVIDIA Carmel (ARMv8.2)", "host_processors_per_node": 1, "host_storage_capacity": "32 GB", "host_storage_type": "eMMC 5.1", "hw_notes": "GPU and both DLAs are used in Offline and MultiStream scenarios", "informal_model": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1024, "mlperf_version": 0.5, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/Xavier", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "pycuda 2019.1, pytorch 1.1, torchvision 0.2.2.post3", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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/Xavier", "system_name": "NVIDIA Jetson AGX Xavier", "target_latency (ns)": 0, "target_qps": 33.9236, "task": "object detection", "task2": "object detection", "total_cores": 8, "uid": "e495c4e979bb327c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" }, { "50.00 percentile latency (ns)": 6614445, "90.00 percentile latency (ns)": 6702585, "90th percentile latency (ns)": 6702585, "95.00 percentile latency (ns)": 6723553, "97.00 percentile latency (ns)": 6738359, "99.00 percentile latency (ns)": 6765422, "99.90 percentile latency (ns)": 6815380, "Max latency (ns)": 8012982, "Mean latency (ns)": 6613923, "Min duration satisfied": "Yes", "Min latency (ns)": 6320198, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 149.72, "QPS w/o loadgen overhead": 151.2, "Result is": "VALID", "SUT name": "LWIS_Server", "Scenario": "singlestream", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "GDDR6", "accelerator_model_name": "NVIDIA Tesla T4", "accelerator_on-chip_memories": "", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.90th_percentile_latency_ms": 6.702585, "characteristics.90th_percentile_latency_ns": 6702585.0, "characteristics.90th_percentile_latency_s": 0.006702585, "characteristics.90th_percentile_latency_us": 6702.585, "characteristics.mAP": 20.067, "ck_system": "T4x8", "ck_used": false, "cooling": "", "dataset": "COCO 2017 (300x300)", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md", "dim_x_default": "characteristics.90th_percentile_latency_ms", "dim_y_default": "characteristics.mAP", "dim_y_maximize": true, "division": "closed", "formal_model": "ssd-mobilenet", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorRT 6.0, CUDA 10.1, cuDNN 7.6.3, libjemalloc2, cub 1.8.0, tensorrt-laboratory mlperf branch", "host_memory_capacity": "768 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 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": "ssd-large", "input_data_types": "int8", "key.accuracy": "characteristics.mAP", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1024, "mlperf_version": 0.5, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/NVIDIA/results/T4x8", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.3", "other_software_stack": "docker 18.09.2, python 3.6.8,gcc 5.5.0,onnx 1.3.0, tensorflow 1.13.1, pytorch 1.1.0, torchvision 0.3.0, pycuda 2019.1, sacrebleu 1.3.3, SimpleJSON, OpenCV 4.1.1", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 64, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "N", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "resnet34-ssd1200.pytorch", "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", "system_name": "Supermicro 4029GP-TRT-OTO-28 8xT4", "target_latency (ns)": 0, "target_qps": 114.742, "task": "object detection", "task2": "object detection", "total_cores": 56, "uid": "4e164abdb74c5dfc", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "quantization, affine fusion" } ]