[ { "50.00 percentile latency (ns)": 37737736028, "90.00 percentile latency (ns)": 67834325793, "95.00 percentile latency (ns)": 71577929402, "97.00 percentile latency (ns)": 73138018550, "99.00 percentile latency (ns)": 74597865795, "99.90 percentile latency (ns)": 75831895889, "Max latency (ns)": 75831895889, "Mean latency (ns)": 37785125138, "Min duration satisfied": "Yes", "Min latency (ns)": 250823919, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "PySUT", "Samples per second": 1218.48, "Scenario": "offline", "accelerator_frequency": "-", "accelerator_host_interconnect": "-", "accelerator_interconnect": "-", "accelerator_interconnect_topology": "-", "accelerator_memory_capacity": "4GB", "accelerator_memory_configuration": "none", "accelerator_model_name": "Centaur Integrated AI Coprocessor", "accelerator_on-chip_memories": "-", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.144, "characteristics.good": 38072, "characteristics.samples_per_second": 1218.48, "characteristics.samples_per_second.normalized_per_core": 1218.48, "characteristics.samples_per_second.normalized_per_processor": 1218.48, "characteristics.total": 50000, "ck_system": "0", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorFlow commit f5ce1c00d4397875ff3d706881bd46430f4a9667 and f6fbfe013898dc16bd35ba380387ff02d0275ac3 + custom patches", "host_memory_capacity": "32GB", "host_processor_core_count": 120, "host_processor_frequency": "2.5GHz", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "120GB", "host_storage_type": "SSD", "informal_model": "resnet", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/CentaurTechnology/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/CentaurTechnology/results/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": "Centaur ML Library Build 1223 and 1220; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "none", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 92400, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "preview", "submitter": "CentaurTechnology", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/CentaurTechnology", "sw_notes": "The same hardware system is used for all mobilenet, resnet, ssd-small, and gnmt sumbissions, but the software is slightly different for gnmt. Submissions for mobilenet, resnet, and ssd-small use TF commit f5ce1c00d4397875ff3d706881bd46430f4a9667 + custom patches to TF-Lite, Centaur ML Library Build 1223, and host_processor_frequency set to 2.5GHz. Submissions for gnmt use TF commit f6fbfe013898dc16bd35ba380387ff02d0275ac3 + custom patches for compiler/graph matching, Centaur ML Library Build 1220, and host_processor_frequency set at runtime via software to 2.3GHz.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/0", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 1400, "task": "image classification", "task2": "image classification", "total_cores": 120, "uid": "c8c6646dd52c22d2", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "TF-Lite v1.14" }, { "50.00 percentile latency (ns)": 33046510536, "90.00 percentile latency (ns)": 58981932805, "95.00 percentile latency (ns)": 62234678086, "97.00 percentile latency (ns)": 63513181633, "99.00 percentile latency (ns)": 64811035480, "99.90 percentile latency (ns)": 65472831372, "Max latency (ns)": 65537564292, "Mean latency (ns)": 33047737508, "Min duration satisfied": "Yes", "Min latency (ns)": 638094043, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "PySUT", "Samples per second": 6042.34, "Scenario": "offline", "accelerator_frequency": "-", "accelerator_host_interconnect": "-", "accelerator_interconnect": "-", "accelerator_interconnect_topology": "-", "accelerator_memory_capacity": "4GB", "accelerator_memory_configuration": "none", "accelerator_model_name": "Centaur Integrated AI Coprocessor", "accelerator_on-chip_memories": "-", "accelerators_per_node": 1, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 70.734, "characteristics.good": 35367, "characteristics.samples_per_second": 6042.34, "characteristics.samples_per_second.normalized_per_core": 6042.34, "characteristics.samples_per_second.normalized_per_processor": 6042.34, "characteristics.total": 50000, "ck_system": "0", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "TensorFlow commit f5ce1c00d4397875ff3d706881bd46430f4a9667 and f6fbfe013898dc16bd35ba380387ff02d0275ac3 + custom patches", "host_memory_capacity": "32GB", "host_processor_core_count": 120, "host_processor_frequency": "2.5GHz", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "120GB", "host_storage_type": "SSD", "informal_model": "mobilenet", "input_data_types": "uint8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 1, "normalize_processors": 1, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/CentaurTechnology/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/CentaurTechnology/results/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": "Centaur ML Library Build 1223 and 1220; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 1024, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "none", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 396000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2269307/files/mobilenet_v1_1.0_224_quant.tgz", "status": "preview", "submitter": "CentaurTechnology", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/CentaurTechnology", "sw_notes": "The same hardware system is used for all mobilenet, resnet, ssd-small, and gnmt sumbissions, but the software is slightly different for gnmt. Submissions for mobilenet, resnet, and ssd-small use TF commit f5ce1c00d4397875ff3d706881bd46430f4a9667 + custom patches to TF-Lite, Centaur ML Library Build 1223, and host_processor_frequency set to 2.5GHz. Submissions for gnmt use TF commit f6fbfe013898dc16bd35ba380387ff02d0275ac3 + custom patches for compiler/graph matching, Centaur ML Library Build 1220, and host_processor_frequency set at runtime via software to 2.3GHz.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/0", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 6000, "task": "image classification", "task2": "image classification", "total_cores": 120, "uid": "f5b3dbaac73d14d3", "use_accelerator": true, "weight_data_types": "uint8", "weight_transformations": "TF-Lite v1.14" } ]