[ { "50.00 percentile latency (ns)": 371548802583, "90.00 percentile latency (ns)": 666528707068, "95.00 percentile latency (ns)": 702671957079, "97.00 percentile latency (ns)": 717692984126, "99.00 percentile latency (ns)": 731696600740, "99.90 percentile latency (ns)": 738314644591, "Max latency (ns)": 738430620948, "Mean latency (ns)": 371859979373, "Min duration satisfied": "Yes", "Min latency (ns)": 5451644570, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "PySUT", "Samples per second": 2860.12, "Scenario": "offline", "accelerator_frequency": "200MHz", "accelerator_host_interconnect": "PCIe 3.0 x16", "accelerator_interconnect": "N/A", "accelerator_interconnect_topology": "N/A", "accelerator_memory_capacity": "32GB", "accelerator_memory_configuration": "DDR4-2666", "accelerator_model_name": "Neuchips RecAccel", "accelerator_on-chip_memories": "5MB SRAM per accelerator", "accelerators_per_node": 3, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 2860.12, "characteristics.samples_per_second.normalized_per_core": 953.3733333333333, "characteristics.samples_per_second.normalized_per_processor": 953.3733333333333, "ck_system": "Neuchips-triple-RecAccel", "ck_used": false, "cooling": "Air-cooled", "dataset": "1TB Click Logs", "dataset_link": "", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.AUC", "dim_y_maximize": true, "division": "closed", "formal_model": "dlrm", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "PyTorch", "host_memory_capacity": "256GB", "host_memory_configuration": "8x32GB DDR4-2666", "host_networking": "N/A", "host_networking_topology": "N/A", "host_processor_caches": "24.75 MB IntelR Smart Cache", "host_processor_core_count": 18, "host_processor_frequency": "4.6GHz", "host_processor_interconnect": "N/A", "host_processor_model_name": "Intel Core i9-10980XE", "host_processors_per_node": 1, "host_storage_capacity": "4TB", "host_storage_type": "SSD", "hw_notes": "Terasic DE-10 Pro FPGA Acceleration Board", "informal_model": "dlrm-99", "input_data_types": "Dense: uint8 / Sparse: uint64", "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": 3, "normalize_processors": 3, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Neuchips/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Neuchips/results/Neuchips-triple-RecAccel", "number_of_nodes": 1, "operating_system": "Ubuntu 16.04.7 LTS", "other_software_stack": "python3.6, gcc 5.4.0", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 204800, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "No", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 2112000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "https://zenodo.org/record/4028341/files/weights_int8.tar.gz?download=1", "status": "RDI", "submitter": "Neuchips", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Neuchips", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Neuchips-triple-RecAccel", "system_name": "RecAccel", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 3200, "task": "recommendation", "task2": "recommendation", "total_cores": 18, "uid": "4034ad897464d3c5", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "Quantization and remapping" }, { "50.00 percentile latency (ns)": 418834416569, "90.00 percentile latency (ns)": 750747796446, "95.00 percentile latency (ns)": 791511563971, "97.00 percentile latency (ns)": 808414535154, "99.00 percentile latency (ns)": 825187921437, "99.90 percentile latency (ns)": 832958869314, "Max latency (ns)": 832958869314, "Mean latency (ns)": 419421788326, "Min duration satisfied": "Yes", "Min latency (ns)": 7082618854, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "PySUT", "Samples per second": 2060.13, "Scenario": "offline", "accelerator_frequency": "200MHz", "accelerator_host_interconnect": "PCIe 3.0 x16", "accelerator_interconnect": "N/A", "accelerator_interconnect_topology": "N/A", "accelerator_memory_capacity": "32GB", "accelerator_memory_configuration": "DDR4-2666", "accelerator_model_name": "Neuchips RecAccel", "accelerator_on-chip_memories": "5MB SRAM per accelerator", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 2060.13, "characteristics.samples_per_second.normalized_per_core": 1030.065, "characteristics.samples_per_second.normalized_per_processor": 1030.065, "ck_system": "Neuchips-dual-RecAccel", "ck_used": true, "cooling": "Air-cooled", "dataset": "1TB Click Logs", "dataset_link": "", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.AUC", "dim_y_maximize": true, "division": "closed", "formal_model": "dlrm", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "PyTorch", "host_memory_capacity": "256GB", "host_memory_configuration": "8x32GB DDR4-2666", "host_networking": "N/A", "host_networking_topology": "N/A", "host_processor_caches": "24.75 MB IntelR Smart Cache", "host_processor_core_count": 120, "host_processor_frequency": "4.6GHz", "host_processor_interconnect": "N/A", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 1, "host_storage_capacity": "4TB", "host_storage_type": "SSD", "hw_notes": "Terasic DE-10 Pro FPGA Acceleration Board", "informal_model": "dlrm-99", "input_data_types": "Dense: uint8 / Sparse: uint64", "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": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Neuchips/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Neuchips/results/Neuchips-dual-RecAccel", "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": "python3.6, gcc 5.4.0; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 204800, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "No", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 1716000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "https://zenodo.org/record/4028341/files/weights_int8.tar.gz?download=1", "status": "RDI", "submitter": "Neuchips", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Neuchips", "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/Neuchips-dual-RecAccel", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 2600, "task": "recommendation", "task2": "recommendation", "total_cores": 120, "uid": "792b476e26b73d2f", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "Quantization and remapping" }, { "50.00 percentile latency (ns)": 36142239417, "90.00 percentile latency (ns)": 58874901137, "95.00 percentile latency (ns)": 61957327559, "97.00 percentile latency (ns)": 62997629796, "99.00 percentile latency (ns)": 63801004490, "99.90 percentile latency (ns)": 63801004490, "Max latency (ns)": 63801004490, "Mean latency (ns)": 36052346100, "Min duration satisfied": "Yes", "Min latency (ns)": 9174594403, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "PySUT", "Samples per second": 1758.59, "Scenario": "offline", "accelerator_frequency": "200MHz", "accelerator_host_interconnect": "PCIe 3.0 x16", "accelerator_interconnect": "N/A", "accelerator_interconnect_topology": "N/A", "accelerator_memory_capacity": "32GB", "accelerator_memory_configuration": "DDR4-2666", "accelerator_model_name": "RecAccel-FPGA", "accelerator_on-chip_memories": "5MB SRAM per accelerator", "accelerators_per_node": 2, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 1758.59, "characteristics.samples_per_second.normalized_per_core": 879.295, "characteristics.samples_per_second.normalized_per_processor": 879.295, "ck_system": "Neuchips-dual-RecAccel", "ck_used": false, "cooling": "Air-cooled", "dataset": "1TB Click Logs", "dataset_link": "", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.AUC", "dim_y_maximize": true, "division": "closed", "formal_model": "dlrm", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "PyTorch", "host_memory_capacity": "256GB", "host_memory_configuration": "8x32GB DDR4-2666", "host_networking": "N/A", "host_networking_topology": "N/A", "host_processor_caches": "24.75 MB IntelR Smart Cache", "host_processor_core_count": 18, "host_processor_frequency": "4.6GHz", "host_processor_interconnect": "N/A", "host_processor_model_name": "Intel Core i9-10980XE", "host_processors_per_node": 1, "host_storage_capacity": "4TB", "host_storage_type": "SSD", "hw_notes": "Terasic DE-10 Pro Acceleration Board with Intel S10 FPGA", "informal_model": "dlrm-99", "input_data_types": "Dense: uint8 / Sparse: uint64", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.7, "normalize_cores": 2, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Neuchips/code", "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Neuchips/results/Neuchips-dual-RecAccel", "number_of_nodes": 1, "operating_system": "Ubuntu 16.04.7 LTS", "other_software_stack": "python3.6, gcc 5.4.0", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 204800, "print_timestamps": true, "problem": false, "qsl_rng_seed": 12786827339337101903, "retraining": "No", "sample_index_rng_seed": 12640797754436136668, "samples_per_query": 112200, "schedule_rng_seed": 3135815929913719677, "starting_weights_filename": "https://zenodo.org/record/4028341/files/weights_int8.tar.gz?download=1", "status": "RDI", "submitter": "Neuchips", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Neuchips", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/Neuchips-dual-RecAccel", "system_name": "RecAccel-FPGA", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 1700, "task": "recommendation", "task2": "recommendation", "total_cores": 18, "uid": "79fd3a47a36884a9", "use_accelerator": true, "weight_data_types": "int8", "weight_transformations": "Quantization and remapping" } ]