[ { "50.00 percentile latency (ns)": 322296393122, "90.00 percentile latency (ns)": 579184241874, "95.00 percentile latency (ns)": 611216376395, "97.00 percentile latency (ns)": 624023345975, "99.00 percentile latency (ns)": 636843331341, "99.90 percentile latency (ns)": 642612067330, "Max latency (ns)": 643252762488, "Mean latency (ns)": 70577959476, "Min duration satisfied": "Yes", "Min latency (ns)": 8748366, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 342011, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "16x PCIe 8 lane Gen4", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "LPDDR4x", "accelerator_model_name": "QUALCOMM Cloud AI 100 PCIe HHHL", "accelerator_on-chip_memories": "144 MB", "accelerators_per_node": 16, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.99, "characteristics.good": 37995, "characteristics.power": 1844.4849144634527, "characteristics.power.normalized_per_core": 115.2803071539658, "characteristics.power.normalized_per_processor": 115.2803071539658, "characteristics.samples_per_second": 342011, "characteristics.samples_per_second.normalized_per_core": 21375.6875, "characteristics.samples_per_second.normalized_per_processor": 21375.6875, "characteristics.total": 50000, "ck_system": "g292_z43_q16-qaic-v1.5.6-aic100", "ck_used": true, "cooling": "air", "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": "Qualcomm Cloud AI SDK v1.5.6", "host_memory_capacity": "16x 64 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "L1d: 32K; L1i: 32K; L2: 512K; L3: 256M", "host_processor_core_count": 64, "host_processor_frequency": "2000 MHz", "host_processor_interconnect": "1x 16-20GT/s xGMI2 (?)", "host_processor_model_name": "AMD EPYC 7713 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1 TB", "host_storage_type": "1x SSD", "hw_notes": "With 75W Accelerator TDP constraints", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 16, "normalize_processors": 16, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Qualcomm/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Qualcomm/results/g292_z43_q16-qaic-v1.5.6-aic100", "number_of_nodes": 1, "operating_system": "CentOS 7 (Linux kernel: 5.4.1-1.el7.elrepo.x86_64 #1 SMP Fri Nov 29 10:21:13 EST 2019 x86_64 x86_64 x86_64 GNU/Linux)", "other_software_stack": "Docker 20.10.7; Collective Knowledge v1.55.5", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 219999780, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Qualcomm", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Qualcomm", "sw_notes": "Powered by Collective Knowledge v1.55.5", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/g292_z43_q16-qaic-v1.5.6-aic100", "system_name": "Gigabyte G292-Z43 16x QAIC100", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 333333, "task": "image classification", "task2": "image classification", "total_cores": 128, "uid": "24c34ba840ec327c", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "see calibration.md" }, { "50.00 percentile latency (ns)": 325162026746, "90.00 percentile latency (ns)": 584991725262, "95.00 percentile latency (ns)": 617495253040, "97.00 percentile latency (ns)": 630474572026, "99.00 percentile latency (ns)": 643499879844, "99.90 percentile latency (ns)": 649347486393, "Max latency (ns)": 649997674896, "Mean latency (ns)": 157351572123, "Min duration satisfied": "Yes", "Min latency (ns)": 3083162, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 169231, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "8x PCIe 8 lane Gen4", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "LPDDR4x", "accelerator_model_name": "QUALCOMM Cloud AI 100 PCIe HHHL", "accelerator_on-chip_memories": "144 MB", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.99, "characteristics.good": 37995, "characteristics.power": 857.2938461538471, "characteristics.power.normalized_per_core": 107.16173076923089, "characteristics.power.normalized_per_processor": 107.16173076923089, "characteristics.samples_per_second": 169231, "characteristics.samples_per_second.normalized_per_core": 21153.875, "characteristics.samples_per_second.normalized_per_processor": 21153.875, "characteristics.total": 50000, "ck_system": "r282_z93_q8-qaic-v1.5.6-aic100", "ck_used": true, "cooling": "air", "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": "Qualcomm Cloud AI SDK v1.5.6", "host_memory_capacity": "8x 64 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "L1d: 32K; L1i: 32K; L2: 512K; L3: 16384K", "host_processor_core_count": 16, "host_processor_frequency": "2800 MHz", "host_processor_interconnect": "1x 16-20GT/s xGMI2", "host_processor_model_name": "AMD EPYC 7282 16-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1 TB", "host_storage_type": "1x SSD", "hw_notes": "With 75W Accelerator TDP constraints. 3x QAIC100 on riser CRS2033; 3x QAIC100 on riser CRS2033; 2x QAIC100 on riser CRS2026", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Qualcomm/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Qualcomm/results/r282_z93_q8-qaic-v1.5.6-aic100", "number_of_nodes": 1, "operating_system": "CentOS 7 (Linux kernel: 5.4.1-1.el7.elrepo.x86_64 #1 SMP Fri Nov 29 10:21:13 EST 2019 x86_64 x86_64 x86_64 GNU/Linux)", "other_software_stack": "Docker 20.10.7; Collective Knowledge v1.55.5", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 109999560, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Qualcomm", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Qualcomm", "sw_notes": "Powered by Collective Knowledge v1.55.5", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/r282_z93_q8-qaic-v1.5.6-aic100", "system_name": "Gigabyte R282-Z93 8x QAIC100", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 166666, "task": "image classification", "task2": "image classification", "total_cores": 32, "uid": "d634e3e8481cc1ae", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "see calibration.md" }, { "50.00 percentile latency (ns)": 322708757934, "90.00 percentile latency (ns)": 591963818083, "95.00 percentile latency (ns)": 625711131028, "97.00 percentile latency (ns)": 639218831720, "99.00 percentile latency (ns)": 652737225042, "99.90 percentile latency (ns)": 658818719087, "Max latency (ns)": 659491153403, "Mean latency (ns)": 45571906351, "Min duration satisfied": "Yes", "Min latency (ns)": 2476850, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 100077, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "5x PCIe 8 lane Gen4", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "16 GB", "accelerator_memory_configuration": "LPDDR4x", "accelerator_model_name": "QUALCOMM Cloud AI 100 PCIe HHHL", "accelerator_on-chip_memories": "144 MB", "accelerators_per_node": 5, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.accuracy": 75.928, "characteristics.good": 37964, "characteristics.power": 561.9439393939398, "characteristics.power.normalized_per_core": 112.38878787878795, "characteristics.power.normalized_per_processor": 112.38878787878795, "characteristics.samples_per_second": 100077, "characteristics.samples_per_second.normalized_per_core": 20015.4, "characteristics.samples_per_second.normalized_per_processor": 20015.4, "characteristics.total": 50000, "ck_system": "r282_z93_q5-qaic-v1.3.14-aic100", "ck_used": true, "cooling": "air", "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": "Qualcomm Cloud AI SDK v1.3.14", "host_memory_capacity": "4x 32 GB", "host_memory_configuration": "", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "L1d: 32K; L1i: 32K; L2: 512K; L3: 16384K", "host_processor_core_count": 120, "host_processor_frequency": "2800 MHz", "host_processor_interconnect": "1x 16-20GT/s xGMI2", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "1 TB", "host_storage_type": "1x Seagate ST1000NM0045 HDD", "hw_notes": "With 75W Accelerator TDP constraints", "informal_model": "resnet50", "input_data_types": "int8", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "normalize_cores": 5, "normalize_processors": 5, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Qualcomm/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Qualcomm/results/r282_z93_q5-qaic-v1.3.14-aic100", "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": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 1024, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "No", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 66000000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Qualcomm", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Qualcomm", "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/r282_z93_q5-qaic-v1.3.14-aic100", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 100000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "bb96fdb514ea6af7", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "see calibration/" } ]