[ { "50.00 percentile latency (ns)": 10585672, "90.00 percentile latency (ns)": 11800417, "95.00 percentile latency (ns)": 12553059, "97.00 percentile latency (ns)": 13111153, "99.00 percentile latency (ns)": 14895845, "99.90 percentile latency (ns)": 59958256, "Completed samples per second": 4021.32, "Max latency (ns)": 69583426, "Mean latency (ns)": 11053800, "Min duration satisfied": "Yes", "Min latency (ns)": 10035072, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "CSUT", "Scenario": "server", "Scheduled samples per second": 4021.94, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.124, "characteristics.good": 38062, "characteristics.scheduled_queries_per_second": 4021.94, "characteristics.scheduled_queries_per_second.normalized_per_core": 8.379041666666668, "characteristics.scheduled_queries_per_second.normalized_per_processor": 1005.485, "characteristics.total": 50000, "ck_system": "clx-8255c-4s_pytorch-caffe2", "ck_used": true, "compilers": "gcc version 7.3.1 ", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "filesystem": "xfs", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "PyTorch", "hardware": "1-node 4x Intel(R) Xeon(R) Platinum 8255C processor", "host_memory_capacity": "768GB", "host_networking": "NA", "host_networking_topology": "NA", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_core_count": 24, "host_processors_frequency": "2.50GHz", "host_processors_interconnect": "UPI", "host_processors_per_node": 4, "host_processors_thread_count": 96, "host_storage_capacity": "422GB", "host_storage_type": "SSD", "informal_model": "resnet50", "input_data_types": "We transform the fp32 input data to int8 data.", "interconnect": "NA", "key.accuracy": "characteristics.accuracy", "libraries": "MKL (v2019.4.243)", "math_libraries": "MKL (2019.4.243)", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "ml_libraries": "Deep Learning Math Kernel Research(Euler) v0.0.3", "mlperf_version": 0.5, "nodes": [ { "accelerator": "NA", "cpu": "Intel(R) Xeon(R) Platinum 8255C CPU @ 2.50GHz", "cpu_accel_interconnect": "NA", "network_card": "NA", "notes": "NA", "num_accelerators": 0, "num_cores": 96, "num_network_cards": 1, "num_nodes": 1, "num_vcpus": "NA", "sys_mem_size": "768GB", "sys_storage_size": "422GB", "sys_storage_type": "SSD" } ], "normalize_cores": 480, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Tencent/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Tencent/results/clx-8255c-4s_pytorch-caffe2", "notes": "none", "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": 50000, "power": "none", "print_timestamps": true, "problem": false, "python_version": "3.6.8", "qsl_rng_seed": 3133965575612453542, "retraining": "We don’t retrain the model weight.", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/2541184/files/resnet50_v1.onnx and the transformed int8 weight named init_net_int8_euler.pb", "status": "preview", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Tencent", "sumitter": "Tencent_Corp", "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/clx-8255c-4s_pytorch-caffe2", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 15000000, "target_qps": 4020, "task": "image classification", "task2": "image classification", "total_cores": 480, "uid": "a06ec841a6b6e05a", "use_accelerator": false, "weight_data_types": "We transform the original fp32 weight to int8 weight.", "weight_transformations": "We transform the weight from fp32 datatype in onnx file to int8 datatype in pb file." }, { "50.00 percentile latency (ns)": 3982518, "90.00 percentile latency (ns)": 4203326, "95.00 percentile latency (ns)": 4361139, "97.00 percentile latency (ns)": 4536427, "99.00 percentile latency (ns)": 6796804, "99.90 percentile latency (ns)": 53762757, "Completed samples per second": 21927.82, "Max latency (ns)": 57985580, "Mean latency (ns)": 4311342, "Min duration satisfied": "Yes", "Min latency (ns)": 3603910, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "CSUT", "Scenario": "server", "Scheduled samples per second": 21929.19, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 70.408, "characteristics.good": 35204, "characteristics.scheduled_queries_per_second": 21929.19, "characteristics.scheduled_queries_per_second.normalized_per_core": 45.6858125, "characteristics.scheduled_queries_per_second.normalized_per_processor": 5482.2975, "characteristics.total": 50000, "ck_system": "clx-8255c-4s_pytorch-caffe2", "ck_used": true, "compilers": "gcc version 7.3.1 ", "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.scheduled_queries_per_second", "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "filesystem": "xfs", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "PyTorch", "hardware": "1-node 4x Intel(R) Xeon(R) Platinum 8255C processor", "host_memory_capacity": "768GB", "host_networking": "NA", "host_networking_topology": "NA", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_core_count": 24, "host_processors_frequency": "2.50GHz", "host_processors_interconnect": "UPI", "host_processors_per_node": 4, "host_processors_thread_count": 96, "host_storage_capacity": "422GB", "host_storage_type": "SSD", "informal_model": "mobilenet", "input_data_types": "We transform the fp32 input data to int8 data.", "interconnect": "NA", "key.accuracy": "characteristics.accuracy", "libraries": "MKL (v2019.4.243)", "math_libraries": "MKL (2019.4.243)", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 270336, "ml_libraries": "Deep Learning Math Kernel Research(Euler) v0.0.3", "mlperf_version": 0.5, "nodes": [ { "accelerator": "NA", "cpu": "Intel(R) Xeon(R) Platinum 8255C CPU @ 2.50GHz", "cpu_accel_interconnect": "NA", "network_card": "NA", "notes": "NA", "num_accelerators": 0, "num_cores": 96, "num_network_cards": 1, "num_nodes": 1, "num_vcpus": "NA", "sys_mem_size": "768GB", "sys_storage_size": "422GB", "sys_storage_type": "SSD" } ], "normalize_cores": 480, "normalize_processors": 4, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Tencent/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Tencent/results/clx-8255c-4s_pytorch-caffe2", "notes": "none", "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": 50000, "power": "none", "print_timestamps": true, "problem": false, "python_version": "3.6.8", "qsl_rng_seed": 3133965575612453542, "retraining": "We don’t retrain the model weight.", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/2541184/files/moblienet_v1.onnx and the transformed int8 weight named init_net_int8_euler.pb", "status": "preview", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Tencent", "sumitter": "Tencent_Corp", "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/clx-8255c-4s_pytorch-caffe2", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 10000000, "target_qps": 21900, "task": "image classification", "task2": "image classification", "total_cores": 480, "uid": "4a6ba069ef72e27c", "use_accelerator": false, "weight_data_types": "We transform the original fp32 weight to int8 weight.", "weight_transformations": "We transform the weight from fp32 datatype in onnx file to int8 datatype in pb file." } ]