[ { "50.00 percentile latency (ns)": 18765265, "90.00 percentile latency (ns)": 19556592, "90th percentile latency (ns)": 19556592, "95.00 percentile latency (ns)": 19710555, "97.00 percentile latency (ns)": 19787665, "99.00 percentile latency (ns)": 19991368, "99.90 percentile latency (ns)": 20386031, "Max latency (ns)": 20600586, "Mean latency (ns)": 18908408, "Min duration satisfied": "Yes", "Min latency (ns)": 18170250, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 52.84, "QPS w/o loadgen overhead": 52.89, "Result is": "VALID", "SUT name": "TFLite_SUT", "Scenario": "singlestream", "accelerator_frequency": "-", "accelerator_host_interconnect": "-", "accelerator_interconnect": "-", "accelerator_interconnect_topology": "-", "accelerator_memory_capacity": "-", "accelerator_memory_configuration": "-", "accelerator_model_name": "Coral USB Accelerator", "accelerator_on-chip_memories": "-", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 19.556592, "characteristics.90th_percentile_latency_ns": 19556592.0, "characteristics.90th_percentile_latency_s": 0.019556592, "characteristics.90th_percentile_latency_us": 19556.592, "characteristics.mAP": 17.93, "ck_system": "rpi4coral-tflite-for.coral-edgetpu", "ck_used": true, "cooling": "MakerHawk Pi-FAN DC Brushless Cooling Fan LD3007MS", "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": "open", "formal_model": "ssd-mobilenet-v1-edgetpu", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TFLite for Coral (EdgeTPU)", "host_memory_capacity": "4 GiB", "host_memory_configuration": "-", "host_networking": "-", "host_networking_topology": "-", "host_processor_caches": "L1I$ 128=4x32 KiB, L1D$ 128=4x32 KiB, L2$ 1 MiB", "host_processor_core_count": 120, "host_processor_frequency": "1500 MHz", "host_processor_interconnect": "-", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "128 GiB", "host_storage_type": "SanDisk Extreme Pro microSD", "hw_notes": "https://www.raspberrypi.org/products/raspberry-pi-4-model-b/specifications/", "informal_model": "ssd-mobilenet-v1-edgetpu", "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": 0.7, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/open/dividiti/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/open/dividiti/results/rpi4coral-tflite-for.coral-edgetpu", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "GCC 7.5.0; Python 3.7.10", "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": 12786827339337101903, "retraining": "no", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "https://github.com/google-coral/edgetpu/raw/master/test_data/ssd_mobilenet_v1_coco_quant_postprocess_edgetpu.tflite", "status": "available", "submitter": "dividiti", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/dividiti", "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/rpi4coral-tflite-for.coral-edgetpu", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 200, "task": "object detection", "task2": "object detection", "total_cores": 120, "uid": "a61e067867cb44cf", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "TF -> EdgeTPU" }, { "50.00 percentile latency (ns)": 25113057, "90.00 percentile latency (ns)": 26007643, "90th percentile latency (ns)": 26007643, "95.00 percentile latency (ns)": 26142292, "97.00 percentile latency (ns)": 26239938, "99.00 percentile latency (ns)": 26485511, "99.90 percentile latency (ns)": 27349431, "Max latency (ns)": 29561564, "Mean latency (ns)": 25127141, "Min duration satisfied": "Yes", "Min latency (ns)": 22526870, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 39.77, "QPS w/o loadgen overhead": 39.8, "Result is": "VALID", "SUT name": "TFLite_SUT", "Scenario": "singlestream", "accelerator_frequency": "-", "accelerator_host_interconnect": "-", "accelerator_interconnect": "-", "accelerator_interconnect_topology": "-", "accelerator_memory_capacity": "-", "accelerator_memory_configuration": "-", "accelerator_model_name": "Coral USB Accelerator", "accelerator_on-chip_memories": "-", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 26.007643, "characteristics.90th_percentile_latency_ns": 26007643.0, "characteristics.90th_percentile_latency_s": 0.026007643, "characteristics.90th_percentile_latency_us": 26007.643, "characteristics.mAP": 22.849, "ck_system": "rpi4coral-tflite-for.coral-edgetpu", "ck_used": true, "cooling": "MakerHawk Pi-FAN DC Brushless Cooling Fan LD3007MS", "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": "open", "formal_model": "ssd-mobilenet-v2-edgetpu", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TFLite for Coral (EdgeTPU)", "host_memory_capacity": "4 GiB", "host_memory_configuration": "-", "host_networking": "-", "host_networking_topology": "-", "host_processor_caches": "L1I$ 128=4x32 KiB, L1D$ 128=4x32 KiB, L2$ 1 MiB", "host_processor_core_count": 120, "host_processor_frequency": "1500 MHz", "host_processor_interconnect": "-", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "128 GiB", "host_storage_type": "SanDisk Extreme Pro microSD", "hw_notes": "https://www.raspberrypi.org/products/raspberry-pi-4-model-b/specifications/", "informal_model": "ssd-mobilenet-v2-edgetpu", "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": 0.7, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/open/dividiti/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/open/dividiti/results/rpi4coral-tflite-for.coral-edgetpu", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "GCC 7.5.0; Python 3.7.10", "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": 12786827339337101903, "retraining": "no", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "https://github.com/google-coral/edgetpu/raw/master/test_data/ssd_mobilenet_v2_coco_quant_postprocess_edgetpu.tflite", "status": "available", "submitter": "dividiti", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/dividiti", "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/rpi4coral-tflite-for.coral-edgetpu", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 200, "task": "object detection", "task2": "object detection", "total_cores": 120, "uid": "c44ca8cd1bc7baa5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "TF -> EdgeTPU" }, { "50.00 percentile latency (ns)": 1945688, "90.00 percentile latency (ns)": 2253701, "90th percentile latency (ns)": 2253701, "95.00 percentile latency (ns)": 4691602, "97.00 percentile latency (ns)": 11531813, "99.00 percentile latency (ns)": 16070320, "99.90 percentile latency (ns)": 18073641, "Max latency (ns)": 100457179, "Mean latency (ns)": 2506984, "Min duration satisfied": "Yes", "Min latency (ns)": 1846323, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 395.26, "QPS w/o loadgen overhead": 398.89, "Result is": "VALID", "SUT name": "PySUT", "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, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 2.253701, "characteristics.90th_percentile_latency_ns": 2253701.0, "characteristics.90th_percentile_latency_s": 0.002253701, "characteristics.90th_percentile_latency_us": 2253.701, "characteristics.mAP": 22.931, "ck_system": "xavier-tensorrt-v6.0", "ck_used": true, "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 v6.0", "host_memory_capacity": "32 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": 120, "host_processor_frequency": "2265.6 MHz", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "400 GB", "host_storage_type": "SanDisk Extreme microSD", "hw_notes": "DLAs are not used in the Offline and MultiStream scenarios (GPU only)", "informal_model": "ssd-mobilenet", "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": 0.7, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/dividiti/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/dividiti/results/xavier-tensorrt-v6.0", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "GCC 7.5.0; Python 3.7.10", "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": 12786827339337101903, "retraining": "no", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "https://zenodo.org/record/3716059/files/ssd-small-SingleStream-gpu-b1-int8-linear.plan", "status": "available", "submitter": "dividiti", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/dividiti", "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/xavier-tensorrt-v6.0", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 1000, "task": "object detection", "task2": "object detection", "total_cores": 120, "uid": "4fc2142e78686021", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "TF -> TensorRT" }, { "50.00 percentile latency (ns)": 153972489, "90.00 percentile latency (ns)": 170988464, "90th percentile latency (ns)": 170988464, "95.00 percentile latency (ns)": 176757587, "97.00 percentile latency (ns)": 182045405, "99.00 percentile latency (ns)": 194405530, "99.90 percentile latency (ns)": 275965551, "Max latency (ns)": 282953286, "Mean latency (ns)": 159590734, "Min duration satisfied": "Yes", "Min latency (ns)": 150902454, "Min queries satisfied": "Yes", "Mode": "Performance", "QPS w/ loadgen overhead": 6.27, "QPS w/o loadgen overhead": 6.27, "Result is": "VALID", "SUT name": "TFLite_SUT", "Scenario": "singlestream", "accelerator_frequency": "-", "accelerator_host_interconnect": "-", "accelerator_interconnect": "-", "accelerator_interconnect_topology": "-", "accelerator_memory_capacity": "-", "accelerator_memory_configuration": "-", "accelerator_model_name": "-", "accelerator_on-chip_memories": "-", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.90th_percentile_latency_ms": 170.988464, "characteristics.90th_percentile_latency_ns": 170988464.0, "characteristics.90th_percentile_latency_s": 0.170988464, "characteristics.90th_percentile_latency_us": 170988.464, "characteristics.mAP": 22.35, "ck_system": "rpi4coral-tflite-v2.2.0-ruy", "ck_used": true, "cooling": "MakerHawk Pi-FAN DC Brushless Cooling Fan LD3007MS", "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": "TFLite v2.2.0 (ruy)", "host_memory_capacity": "4 GiB", "host_memory_configuration": "-", "host_networking": "-", "host_networking_topology": "-", "host_processor_caches": "L1I$ 128=4x32 KiB, L1D$ 128=4x32 KiB, L2$ 1 MiB", "host_processor_core_count": 4, "host_processor_frequency": "1500 MHz", "host_processor_interconnect": "-", "host_processor_model_name": "Arm Cortex-A72 MP4", "host_processors_per_node": 1, "host_storage_capacity": "128 GiB", "host_storage_type": "SanDisk Extreme Pro microSD", "hw_notes": "https://www.raspberrypi.org/products/raspberry-pi-4-model-b/specifications/", "informal_model": "ssd-mobilenet-non-quantized", "input_data_types": "fp32", "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": 0.7, "normalize_cores": 4, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/dividiti/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/dividiti/results/rpi4coral-tflite-v2.2.0-ruy", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.1; kernel 5.4.0-1018-raspi #20-Ubuntu SMP (Sun Sep 6 05:11:16 UTC 2020)", "other_software_stack": "GCC 9.3.0; Python 3.8.2", "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": 12786827339337101903, "retraining": "no", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 1, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "https://zenodo.org/record/3361502/files/tflite_ssd_mobilenet_v1_coco_2018_01_28.tar.gz", "status": "available", "submitter": "dividiti", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/dividiti", "sw_notes": "Powered by Collective Knowledge v1.15.0", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/rpi4coral-tflite-v2.2.0-ruy", "system_name": "Raspberry Pi 4 with Coral (rpi4coral)", "system_type": "edge", "target_latency (ns)": 0, "target_qps": 10, "task": "object detection", "task2": "object detection", "total_cores": 4, "uid": "2e9ac862bf44bd8f", "use_accelerator": false, "weight_data_types": "fp32", "weight_transformations": "TF -> TFLite" } ]