[ { "50.00 percentile latency (ns)": 7287265, "90.00 percentile latency (ns)": 7859935, "95.00 percentile latency (ns)": 8106133, "97.00 percentile latency (ns)": 8303993, "99.00 percentile latency (ns)": 8759132, "99.90 percentile latency (ns)": 9703943, "Completed samples per second": 21029.07, "Max latency (ns)": 10794909, "Mean latency (ns)": 7349575, "Min duration satisfied": "Yes", "Min latency (ns)": 6508445, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "sut", "Scenario": "server", "Scheduled samples per second": 21031.98, "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.482, "characteristics.good": 38241, "characteristics.scheduled_queries_per_second": 21031.98, "characteristics.scheduled_queries_per_second.normalized_per_core": 2628.9975, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2628.9975, "characteristics.total": 50000, "ck_system": "tpu-rdi-v3.8x1", "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.scheduled_queries_per_second", "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": "TF at the internal CL listed in the code dir", "host_memory_capacity": "383 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "21.42 GB", "host_storage_type": "HDD", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1260000, "mlperf_version": 0.5, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-rdi-v3.8x1", "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": "TF; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "rdi", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-rdi-v3.8x1", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 15000000, "target_qps": 21000, "task": "image classification", "task2": "image classification", "total_cores": 120, "uid": "9f3a64b7450b806b", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "Transpose the weight of the first convolution layer" }, { "50.00 percentile latency (ns)": 9010265, "90.00 percentile latency (ns)": 9608185, "95.00 percentile latency (ns)": 9884772, "97.00 percentile latency (ns)": 10436154, "99.00 percentile latency (ns)": 13716900, "99.90 percentile latency (ns)": 21918252, "Completed samples per second": 16010.21, "Max latency (ns)": 36058606, "Mean latency (ns)": 9168653, "Min duration satisfied": "Yes", "Min latency (ns)": 8084644, "Min queries satisfied": "Yes", "Mode": "Performance", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "sut", "Scenario": "server", "Scheduled samples per second": 16014.29, "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.482, "characteristics.good": 38241, "characteristics.scheduled_queries_per_second": 16014.29, "characteristics.scheduled_queries_per_second.normalized_per_core": 2001.78625, "characteristics.scheduled_queries_per_second.normalized_per_processor": 2001.78625, "characteristics.total": 50000, "ck_system": "tpu-v3.8x1", "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.scheduled_queries_per_second", "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": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 960000, "mlperf_version": 0.5, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x1", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 1, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x1", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 15000000, "target_qps": 16000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "498daf890b267b17", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "Transpose the weight of the first convolution layer" } ]