[ { "50.00 percentile latency (ns)": 329997951384, "90.00 percentile latency (ns)": 593911805292, "95.00 percentile latency (ns)": 626883979584, "97.00 percentile latency (ns)": 640071877517, "99.00 percentile latency (ns)": 653273015086, "99.90 percentile latency (ns)": 659231285527, "Max latency (ns)": 659884201206, "Mean latency (ns)": 329988439096, "Min duration satisfied": "Yes", "Min latency (ns)": 15159483, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 10601.9, "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.power": 1712.0945454545467, "characteristics.power.normalized_per_core": 107.00590909090917, "characteristics.power.normalized_per_processor": 107.00590909090917, "characteristics.samples_per_second": 10601.9, "characteristics.samples_per_second.normalized_per_core": 662.61875, "characteristics.samples_per_second.normalized_per_processor": 662.61875, "ck_system": "g292_z43_q16-qaic-v1.5.9-aic100", "ck_used": true, "cooling": "air", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "Qualcomm Cloud AI SDK v1.5.9", "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": "bert-99", "input_data_types": "int64", "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-v1.5.9-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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 6996000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/3733868/files/model.ckpt-5474.data-00000-of-00001", "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-v1.5.9-aic100", "system_name": "Gigabyte G292-Z43 16x QAIC100", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 10600, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "396014b617845c6a", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "see calibration.md" }, { "50.00 percentile latency (ns)": 330124369240, "90.00 percentile latency (ns)": 595030550197, "95.00 percentile latency (ns)": 628193746891, "97.00 percentile latency (ns)": 641476377676, "99.00 percentile latency (ns)": 654743102617, "99.90 percentile latency (ns)": 660720679514, "Max latency (ns)": 661385502913, "Mean latency (ns)": 330285411566, "Min duration satisfied": "Yes", "Min latency (ns)": 25744181, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 5508.44, "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.power": 1881.1429652042368, "characteristics.power.normalized_per_core": 117.5714353252648, "characteristics.power.normalized_per_processor": 117.5714353252648, "characteristics.samples_per_second": 5508.44, "characteristics.samples_per_second.normalized_per_core": 344.2775, "characteristics.samples_per_second.normalized_per_processor": 344.2775, "ck_system": "g292_z43_q16-qaic-v1.5.9-aic100", "ck_used": true, "cooling": "air", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "Qualcomm Cloud AI SDK v1.5.9", "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": "bert-99.9", "input_data_types": "int64", "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-v1.5.9-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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 3643200, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/3733868/files/model.ckpt-5474.data-00000-of-00001", "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-v1.5.9-aic100", "system_name": "Gigabyte G292-Z43 16x QAIC100", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 5520, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "5a8446d9f45276ec", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "fp32 -> fp16" }, { "50.00 percentile latency (ns)": 329191959919, "90.00 percentile latency (ns)": 593664133143, "95.00 percentile latency (ns)": 626707595889, "97.00 percentile latency (ns)": 639928739602, "99.00 percentile latency (ns)": 653145489597, "99.90 percentile latency (ns)": 659096294657, "Max latency (ns)": 659761729622, "Mean latency (ns)": 329423860565, "Min duration satisfied": "Yes", "Min latency (ns)": 13999075, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 5202.88, "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.power": 776.2986363636365, "characteristics.power.normalized_per_core": 97.03732954545457, "characteristics.power.normalized_per_processor": 97.03732954545457, "characteristics.samples_per_second": 5202.88, "characteristics.samples_per_second.normalized_per_core": 650.36, "characteristics.samples_per_second.normalized_per_processor": 650.36, "ck_system": "r282_z93_q8-qaic-v1.5.9-aic100", "ck_used": true, "cooling": "air", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "Qualcomm Cloud AI SDK v1.5.9", "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": "bert-99", "input_data_types": "int64", "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-v1.5.9-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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 3432660, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/3733868/files/model.ckpt-5474.data-00000-of-00001", "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-v1.5.9-aic100", "system_name": "Gigabyte R282-Z93 8x QAIC100", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 5201, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "bacb372b49c9f34c", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "see calibration.md" }, { "50.00 percentile latency (ns)": 332488719747, "90.00 percentile latency (ns)": 601529328415, "95.00 percentile latency (ns)": 635196398783, "97.00 percentile latency (ns)": 648649598813, "99.00 percentile latency (ns)": 662140551735, "99.90 percentile latency (ns)": 668214597872, "Max latency (ns)": 668883050724, "Mean latency (ns)": 333073246362, "Min duration satisfied": "Yes", "Min latency (ns)": 26110308, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "QAIC_SUT", "Samples per second": 2664.14, "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.power": 841.4444610778444, "characteristics.power.normalized_per_core": 105.18055763473055, "characteristics.power.normalized_per_processor": 105.18055763473055, "characteristics.samples_per_second": 2664.14, "characteristics.samples_per_second.normalized_per_core": 333.0175, "characteristics.samples_per_second.normalized_per_processor": 333.0175, "ck_system": "r282_z93_q8-qaic-v1.5.9-aic100", "ck_used": true, "cooling": "air", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.9, "formal_model_link": "", "framework": "Qualcomm Cloud AI SDK v1.5.9", "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": "bert-99.9", "input_data_types": "int64", "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-v1.5.9-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": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 1782000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/3733868/files/model.ckpt-5474.data-00000-of-00001", "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-v1.5.9-aic100", "system_name": "Gigabyte R282-Z93 8x QAIC100", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 2700, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "78005cb479ba0207", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "fp32 -> fp16" } ]