[
  {
    "50.00 percentile latency (ns)": 72755604281,
    "90.00 percentile latency (ns)": 126164840098,
    "95.00 percentile latency (ns)": 133005338121,
    "97.00 percentile latency (ns)": 135617501767,
    "99.00 percentile latency (ns)": 138219823855,
    "99.90 percentile latency (ns)": 140018878020,
    "Max latency (ns)": 140886034441,
    "Mean latency (ns)": 72780421979,
    "Min duration satisfied": "Yes",
    "Min latency (ns)": 2705463056,
    "Min queries satisfied": "Yes",
    "Mode": "Performance",
    "Result is": "VALID",
    "SUT name": "PySUT",
    "Samples per second": 2623.4,
    "Scenario": "offline",
    "accelerator_memory_capacity": "16 GB per accelerator",
    "accelerator_memory_configuration": "HBM",
    "accelerator_model_name": "tpu-v3.8",
    "accelerators_per_node": 32,
    "accuracy_log_probability": 0,
    "accuracy_log_rng_seed": 0,
    "characteristics.mAP": 20.779,
    "characteristics.samples_per_second": 2623.4,
    "characteristics.samples_per_second.normalized_per_core": 81.98125,
    "characteristics.samples_per_second.normalized_per_processor": 81.98125,
    "ck_system": "tpu-v3.8x4",
    "ck_used": true,
    "dataset": "COCO 2017 (300x300)",
    "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md",
    "dim_x_default": "characteristics.samples_per_second",
    "dim_x_maximize": true,
    "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": "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": "ssd-large",
    "input_data_types": "bfloat16",
    "key.accuracy": "characteristics.mAP",
    "max_async_queries": 1,
    "max_duration (ms)": 0,
    "max_query_count": 0,
    "min_duration (ms)": 120000,
    "min_query_count": 1,
    "mlperf_version": 0.5,
    "normalize_cores": 32,
    "normalize_processors": 32,
    "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-v3.8x4",
    "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": 64,
    "print_timestamps": true,
    "problem": false,
    "qsl_rng_seed": 3133965575612453542,
    "retraining": "None",
    "sample_index_rng_seed": 665484352860916858,
    "samples_per_query": 369600,
    "schedule_rng_seed": 3622009729038561421,
    "starting_weights_filename": "https://zenodo.org/record/3345892/files/tf_ssd_resnet34_22.1.zip",
    "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-v3.8x4",
    "system_name": "Microsoft Corporation 7.0 (Virtual Machine)",
    "target_latency (ns)": 0,
    "target_qps": 2800,
    "task": "object detection",
    "task2": "object detection",
    "total_cores": 240,
    "uid": "379105abefa58ff4",
    "use_accelerator": true,
    "weight_data_types": "bfloat16",
    "weight_transformations": "None"
  },
  {
    "50.00 percentile latency (ns)": 72484241430,
    "90.00 percentile latency (ns)": 125966490657,
    "95.00 percentile latency (ns)": 132686085374,
    "97.00 percentile latency (ns)": 135371400160,
    "99.00 percentile latency (ns)": 138037854080,
    "99.90 percentile latency (ns)": 139965159156,
    "Max latency (ns)": 140515138908,
    "Mean latency (ns)": 72469702250,
    "Min duration satisfied": "Yes",
    "Min latency (ns)": 2849506937,
    "Min queries satisfied": "Yes",
    "Mode": "Performance",
    "Result is": "VALID",
    "SUT name": "PySUT",
    "Samples per second": 5260.64,
    "Scenario": "offline",
    "accelerator_memory_capacity": "16 GB per accelerator",
    "accelerator_memory_configuration": "HBM",
    "accelerator_model_name": "tpu-v3.8",
    "accelerators_per_node": 64,
    "accuracy_log_probability": 0,
    "accuracy_log_rng_seed": 0,
    "characteristics.mAP": 20.779,
    "characteristics.samples_per_second": 5260.64,
    "characteristics.samples_per_second.normalized_per_core": 82.1975,
    "characteristics.samples_per_second.normalized_per_processor": 82.1975,
    "ck_system": "tpu-v3.8x8",
    "ck_used": true,
    "dataset": "COCO 2017 (300x300)",
    "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md",
    "dim_x_default": "characteristics.samples_per_second",
    "dim_x_maximize": true,
    "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": "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": "ssd-large",
    "input_data_types": "bfloat16",
    "key.accuracy": "characteristics.mAP",
    "max_async_queries": 1,
    "max_duration (ms)": 0,
    "max_query_count": 0,
    "min_duration (ms)": 120000,
    "min_query_count": 1,
    "mlperf_version": 0.5,
    "normalize_cores": 64,
    "normalize_processors": 64,
    "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-v3.8x8",
    "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": 64,
    "print_timestamps": true,
    "problem": false,
    "qsl_rng_seed": 3133965575612453542,
    "retraining": "None",
    "sample_index_rng_seed": 665484352860916858,
    "samples_per_query": 739200,
    "schedule_rng_seed": 3622009729038561421,
    "starting_weights_filename": "https://zenodo.org/record/3345892/files/tf_ssd_resnet34_22.1.zip",
    "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-v3.8x8",
    "system_name": "Microsoft Corporation 7.0 (Virtual Machine)",
    "target_latency (ns)": 0,
    "target_qps": 5600,
    "task": "object detection",
    "task2": "object detection",
    "total_cores": 240,
    "uid": "ad9dcf39ef543f43",
    "use_accelerator": true,
    "weight_data_types": "bfloat16",
    "weight_transformations": "None"
  },
  {
    "50.00 percentile latency (ns)": 72802835168,
    "90.00 percentile latency (ns)": 126293657114,
    "95.00 percentile latency (ns)": 133015627121,
    "97.00 percentile latency (ns)": 135886092285,
    "99.00 percentile latency (ns)": 138375697415,
    "99.90 percentile latency (ns)": 139708850065,
    "Max latency (ns)": 140272320338,
    "Mean latency (ns)": 72898068290,
    "Min duration satisfied": "Yes",
    "Min latency (ns)": 5145022353,
    "Min queries satisfied": "Yes",
    "Mode": "Performance",
    "Result is": "VALID",
    "SUT name": "PySUT",
    "Samples per second": 1317.44,
    "Scenario": "offline",
    "accelerator_memory_capacity": "16 GB per accelerator",
    "accelerator_memory_configuration": "HBM",
    "accelerator_model_name": "tpu-v3.8",
    "accelerators_per_node": 16,
    "accuracy_log_probability": 0,
    "accuracy_log_rng_seed": 0,
    "characteristics.mAP": 20.779,
    "characteristics.samples_per_second": 1317.44,
    "characteristics.samples_per_second.normalized_per_core": 82.34,
    "characteristics.samples_per_second.normalized_per_processor": 82.34,
    "ck_system": "tpu-v3.8x2",
    "ck_used": true,
    "dataset": "COCO 2017 (300x300)",
    "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md",
    "dim_x_default": "characteristics.samples_per_second",
    "dim_x_maximize": true,
    "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": "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": "ssd-large",
    "input_data_types": "bfloat16",
    "key.accuracy": "characteristics.mAP",
    "max_async_queries": 1,
    "max_duration (ms)": 0,
    "max_query_count": 0,
    "min_duration (ms)": 120000,
    "min_query_count": 1,
    "mlperf_version": 0.5,
    "normalize_cores": 16,
    "normalize_processors": 16,
    "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-v3.8x2",
    "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": 64,
    "print_timestamps": true,
    "problem": false,
    "qsl_rng_seed": 3133965575612453542,
    "retraining": "None",
    "sample_index_rng_seed": 665484352860916858,
    "samples_per_query": 184800,
    "schedule_rng_seed": 3622009729038561421,
    "starting_weights_filename": "https://zenodo.org/record/3345892/files/tf_ssd_resnet34_22.1.zip",
    "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-v3.8x2",
    "system_name": "Microsoft Corporation 7.0 (Virtual Machine)",
    "target_latency (ns)": 0,
    "target_qps": 1400,
    "task": "object detection",
    "task2": "object detection",
    "total_cores": 240,
    "uid": "bbd86024b60b934f",
    "use_accelerator": true,
    "weight_data_types": "bfloat16",
    "weight_transformations": "None"
  },
  {
    "50.00 percentile latency (ns)": 105239089646,
    "90.00 percentile latency (ns)": 183452465700,
    "95.00 percentile latency (ns)": 193999550414,
    "97.00 percentile latency (ns)": 197304916057,
    "99.00 percentile latency (ns)": 201647798178,
    "99.90 percentile latency (ns)": 202921266609,
    "Max latency (ns)": 203239273528,
    "Mean latency (ns)": 104160834745,
    "Min duration satisfied": "Yes",
    "Min latency (ns)": 3545049972,
    "Min queries satisfied": "Yes",
    "Mode": "Performance",
    "Result is": "VALID",
    "SUT name": "PySUT",
    "Samples per second": 7274.18,
    "Scenario": "offline",
    "accelerator_memory_capacity": "16 GB per accelerator",
    "accelerator_memory_configuration": "HBM",
    "accelerator_model_name": "tpu-v3.8",
    "accelerators_per_node": 128,
    "accuracy_log_probability": 0,
    "accuracy_log_rng_seed": 0,
    "characteristics.mAP": 20.779,
    "characteristics.samples_per_second": 7274.18,
    "characteristics.samples_per_second.normalized_per_core": 56.82953125,
    "characteristics.samples_per_second.normalized_per_processor": 56.82953125,
    "ck_system": "tpu-v3.8x16",
    "ck_used": true,
    "dataset": "COCO 2017 (300x300)",
    "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md",
    "dim_x_default": "characteristics.samples_per_second",
    "dim_x_maximize": true,
    "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": "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": "ssd-large",
    "input_data_types": "bfloat16",
    "key.accuracy": "characteristics.mAP",
    "max_async_queries": 1,
    "max_duration (ms)": 0,
    "max_query_count": 0,
    "min_duration (ms)": 120000,
    "min_query_count": 1,
    "mlperf_version": 0.5,
    "normalize_cores": 128,
    "normalize_processors": 128,
    "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-v3.8x16",
    "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": 64,
    "print_timestamps": true,
    "problem": false,
    "qsl_rng_seed": 3133965575612453542,
    "retraining": "None",
    "sample_index_rng_seed": 665484352860916858,
    "samples_per_query": 1478400,
    "schedule_rng_seed": 3622009729038561421,
    "starting_weights_filename": "https://zenodo.org/record/3345892/files/tf_ssd_resnet34_22.1.zip",
    "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-v3.8x16",
    "system_name": "Microsoft Corporation 7.0 (Virtual Machine)",
    "target_latency (ns)": 0,
    "target_qps": 11200,
    "task": "object detection",
    "task2": "object detection",
    "total_cores": 240,
    "uid": "c9ec27d4fe8a7d50",
    "use_accelerator": true,
    "weight_data_types": "bfloat16",
    "weight_transformations": "None"
  },
  {
    "50.00 percentile latency (ns)": 72660527607,
    "90.00 percentile latency (ns)": 126435137802,
    "95.00 percentile latency (ns)": 133205090992,
    "97.00 percentile latency (ns)": 136098673794,
    "99.00 percentile latency (ns)": 138744298548,
    "99.90 percentile latency (ns)": 140759436909,
    "Max latency (ns)": 140759436909,
    "Mean latency (ns)": 73100365181,
    "Min duration satisfied": "Yes",
    "Min latency (ns)": 5721945249,
    "Min queries satisfied": "Yes",
    "Mode": "Performance",
    "Result is": "VALID",
    "SUT name": "PySUT",
    "Samples per second": 656.439,
    "Scenario": "offline",
    "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.mAP": 20.779,
    "characteristics.samples_per_second": 656.439,
    "characteristics.samples_per_second.normalized_per_core": 82.054875,
    "characteristics.samples_per_second.normalized_per_processor": 82.054875,
    "ck_system": "tpu-v3.8x1",
    "ck_used": true,
    "dataset": "COCO 2017 (300x300)",
    "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/coco2017.md",
    "dim_x_default": "characteristics.samples_per_second",
    "dim_x_maximize": true,
    "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": "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": "ssd-large",
    "input_data_types": "bfloat16",
    "key.accuracy": "characteristics.mAP",
    "max_async_queries": 1,
    "max_duration (ms)": 0,
    "max_query_count": 0,
    "min_duration (ms)": 120000,
    "min_query_count": 1,
    "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": 64,
    "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/3345892/files/tf_ssd_resnet34_22.1.zip",
    "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)": 0,
    "target_qps": 700,
    "task": "object detection",
    "task2": "object detection",
    "total_cores": 240,
    "uid": "6202fd47d69f5627",
    "use_accelerator": true,
    "weight_data_types": "bfloat16",
    "weight_transformations": "None"
  }
]