[ { "50.00 percentile latency (ns)": 82410683, "90.00 percentile latency (ns)": 104605237, "95.00 percentile latency (ns)": 107952640, "97.00 percentile latency (ns)": 110125114, "99.00 percentile latency (ns)": 115210368, "99.90 percentile latency (ns)": 127620850, "Completed samples per second": 10302.89, "Max latency (ns)": 142624078, "Mean latency (ns)": 70703898, "Min duration satisfied": "Yes", "Min latency (ns)": 12245191, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "QAIC_SUT", "Scenario": "server", "Scheduled samples per second": 10303.27, "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": 1732.2905000000007, "characteristics.power.normalized_per_core": 108.26815625000005, "characteristics.power.normalized_per_processor": 108.26815625000005, "characteristics.scheduled_queries_per_second": 10303.27, "characteristics.scheduled_queries_per_second.normalized_per_core": 643.954375, "characteristics.scheduled_queries_per_second.normalized_per_processor": 643.954375, "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_y_default": "characteristics.scheduled_queries_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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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": 1, "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)": 130000000, "target_qps": 10301, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "0bb7a7b12ae4fc99", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "see calibration.md" }, { "50.00 percentile latency (ns)": 63629677, "90.00 percentile latency (ns)": 91567939, "95.00 percentile latency (ns)": 96119788, "97.00 percentile latency (ns)": 98444507, "99.00 percentile latency (ns)": 105763129, "99.90 percentile latency (ns)": 219170247, "Completed samples per second": 5102.07, "Max latency (ns)": 269468817, "Mean latency (ns)": 64195054, "Min duration satisfied": "Yes", "Min latency (ns)": 19039667, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "QAIC_SUT", "Scenario": "server", "Scheduled samples per second": 5102.64, "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": 1819.1698333333343, "characteristics.power.normalized_per_core": 113.69811458333339, "characteristics.power.normalized_per_processor": 113.69811458333339, "characteristics.scheduled_queries_per_second": 5102.64, "characteristics.scheduled_queries_per_second.normalized_per_core": 318.915, "characteristics.scheduled_queries_per_second.normalized_per_processor": 318.915, "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_y_default": "characteristics.scheduled_queries_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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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": 1, "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)": 130000000, "target_qps": 5100, "task": "NLP", "task2": "nlp", "total_cores": 128, "uid": "92240b8b4f7e45eb", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "fp32 -> fp16" }, { "50.00 percentile latency (ns)": 25402289, "90.00 percentile latency (ns)": 55459877, "95.00 percentile latency (ns)": 83611998, "97.00 percentile latency (ns)": 94293052, "99.00 percentile latency (ns)": 103438263, "99.90 percentile latency (ns)": 111620150, "Completed samples per second": 4902.7, "Max latency (ns)": 119748023, "Mean latency (ns)": 31794242, "Min duration satisfied": "Yes", "Min latency (ns)": 11900810, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "QAIC_SUT", "Scenario": "server", "Scheduled samples per second": 4902.82, "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": 765.1739999999991, "characteristics.power.normalized_per_core": 95.64674999999988, "characteristics.power.normalized_per_processor": 95.64674999999988, "characteristics.scheduled_queries_per_second": 4902.82, "characteristics.scheduled_queries_per_second.normalized_per_core": 612.8525, "characteristics.scheduled_queries_per_second.normalized_per_processor": 612.8525, "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_y_default": "characteristics.scheduled_queries_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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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": 1, "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)": 130000000, "target_qps": 4901, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "145468046dba4af5", "use_accelerator": true, "weight_data_types": "int8,fp16", "weight_transformations": "see calibration.md" }, { "50.00 percentile latency (ns)": 61379212, "90.00 percentile latency (ns)": 88836863, "95.00 percentile latency (ns)": 94235353, "97.00 percentile latency (ns)": 96645253, "99.00 percentile latency (ns)": 100811006, "99.90 percentile latency (ns)": 115876124, "Completed samples per second": 2249.94, "Max latency (ns)": 214487775, "Mean latency (ns)": 61548649, "Min duration satisfied": "Yes", "Min latency (ns)": 20270944, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Performance constraints satisfied": "Yes", "Result is": "VALID", "SUT name": "QAIC_SUT", "Scenario": "server", "Scheduled samples per second": 2250.17, "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": 763.915973377704, "characteristics.power.normalized_per_core": 95.489496672213, "characteristics.power.normalized_per_processor": 95.489496672213, "characteristics.scheduled_queries_per_second": 2250.17, "characteristics.scheduled_queries_per_second.normalized_per_core": 281.27125, "characteristics.scheduled_queries_per_second.normalized_per_processor": 281.27125, "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_y_default": "characteristics.scheduled_queries_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": 0, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 270336, "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": 1, "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)": 130000000, "target_qps": 2250, "task": "NLP", "task2": "nlp", "total_cores": 32, "uid": "4ad2243c2dc850ea", "use_accelerator": true, "weight_data_types": "fp16", "weight_transformations": "fp32 -> fp16" } ]