nlp-benchmarks

AWS EC2 Amazon Linux 2023 Benchmarking

HTML result view exported from: https://openbenchmarking.org/result/2402089-NE-2402012NE50&sor&gru.

nlp-benchmarksProcessorMotherboardChipsetMemoryDiskNetworkOSKernelCompilerFile-SystemSystem Layerc6i.2xlargem7i-flex.2xlargec7a.2xlargeIntel Xeon Platinum 8375C (4 Cores / 8 Threads)Amazon EC2 c6i.2xlarge (1.0 BIOS)Intel 440FX 82441FX PMC1 x 16GB DDR4-3200MT/s215GB Amazon Elastic Block StoreAmazon ElasticAmazon Linux 20236.1.61-85.141.amzn2023.x86_64 (x86_64)GCC 11.4.1 20230605xfsamazonIntel Xeon Platinum 8488C (4 Cores / 8 Threads)Amazon EC2 m7i-flex.2xlarge (1.0 BIOS)1 x 32GB 4800MT/s6.1.72-96.166.amzn2023.x86_64 (x86_64)AMD EPYC 9R14 (8 Cores)Amazon EC2 c7a.2xlarge (1.0 BIOS)1 x 16GB 4800MT/sOpenBenchmarking.orgKernel Details- Transparent Huge Pages: madviseCompiler Details- --build=x86_64-amazon-linux --disable-libunwind-exceptions --enable-__cxa_atexit --enable-bootstrap --enable-cet --enable-checking=release --enable-gnu-indirect-function --enable-gnu-unique-object --enable-initfini-array --enable-languages=c,c++,fortran,lto --enable-multilib --enable-offload-targets=nvptx-none --enable-plugin --enable-shared --enable-threads=posix --mandir=/usr/share/man --with-arch_32=x86-64 --with-arch_64=x86-64-v2 --with-gcc-major-version-only --with-linker-hash-style=gnu --with-tune=generic --without-cuda-driver Processor Details- c6i.2xlarge: CPU Microcode: 0xd0003a5- m7i-flex.2xlarge: CPU Microcode: 0x2b000571- c7a.2xlarge: CPU Microcode: 0xa10113ePython Details- Python 3.11.6Security Details- c6i.2xlarge: gather_data_sampling: Unknown: Dependent on hypervisor status + itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Mitigation of Clear buffers; SMT Host state unknown + retbleed: Not affected + spec_rstack_overflow: Not affected + spec_store_bypass: Mitigation of SSB disabled via prctl + spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization + spectre_v2: Mitigation of Enhanced IBRS IBPB: conditional RSB filling PBRSB-eIBRS: SW sequence + srbds: Not affected + tsx_async_abort: Not affected - m7i-flex.2xlarge: gather_data_sampling: Not affected + itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: Not affected + spec_rstack_overflow: Not affected + spec_store_bypass: Mitigation of SSB disabled via prctl + spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization + spectre_v2: Mitigation of Enhanced IBRS IBPB: conditional RSB filling PBRSB-eIBRS: SW sequence + srbds: Not affected + tsx_async_abort: Not affected - c7a.2xlarge: gather_data_sampling: Not affected + itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: Not affected + spec_rstack_overflow: Mitigation of safe RET no microcode + spec_store_bypass: Mitigation of SSB disabled via prctl + spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization + spectre_v2: Mitigation of Retpolines IBPB: conditional IBRS_FW STIBP: disabled RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

nlp-benchmarkspytorch: CPU - 1 - ResNet-50pytorch: CPU - 1 - ResNet-152pytorch: CPU - 16 - ResNet-50pytorch: CPU - 32 - ResNet-50pytorch: CPU - 16 - ResNet-152pytorch: CPU - 32 - ResNet-152pytorch: CPU - 1 - Efficientnet_v2_lpytorch: CPU - 16 - Efficientnet_v2_lpytorch: CPU - 32 - Efficientnet_v2_lopenvino: Face Detection FP16 - CPUopenvino: Face Detection FP16-INT8 - CPUopenvino: Machine Translation EN To DE FP16 - CPUnumpy: pybench: Total For Average Test Timesonednn: Convolution Batch Shapes Auto - f32 - CPUonednn: Deconvolution Batch shapes_1d - f32 - CPUonednn: Deconvolution Batch shapes_3d - f32 - CPUonednn: Convolution Batch Shapes Auto - u8s8f32 - CPUonednn: Deconvolution Batch shapes_1d - u8s8f32 - CPUonednn: Deconvolution Batch shapes_3d - u8s8f32 - CPUonednn: Recurrent Neural Network Inference - f32 - CPUonednn: Convolution Batch Shapes Auto - bf16bf16bf16 - CPUonednn: Deconvolution Batch shapes_1d - bf16bf16bf16 - CPUonednn: Deconvolution Batch shapes_3d - bf16bf16bf16 - CPUonednn: Recurrent Neural Network Inference - u8s8f32 - CPUonednn: Recurrent Neural Network Inference - bf16bf16bf16 - CPUopenvino: Face Detection FP16 - CPUopenvino: Face Detection FP16-INT8 - CPUopenvino: Machine Translation EN To DE FP16 - CPUc6i.2xlargem7i-flex.2xlargec7a.2xlarge26.7810.5715.9615.816.366.387.994.064.041.776.5322.26374.9910005.9366912.66188.098036.243912.038361.772862496.8433.192046.698634.71662492.292501.502252.04610.59179.5329.8912.1018.8418.347.367.558.995.385.478.4316.5753.87438.257365.842618.100077.948787.194621.0163271.0032692382.463.689931.763503.110752389.642318.31474.46241.2174.2050.1620.0031.8931.2912.9513.0311.748.758.715.169.7654.51590.108877.351015.014375.103307.746191.094291.264051482.353.282456.902003.123391480.931478.51774.23409.7873.34OpenBenchmarking.org

PyTorch

Device: CPU - Batch Size: 1 - Model: ResNet-50

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 1 - Model: ResNet-50c7a.2xlargem7i-flex.2xlargec6i.2xlarge1122334455SE +/- 0.32, N = 3SE +/- 0.05, N = 3SE +/- 0.13, N = 350.1629.8926.78MIN: 33.27 / MAX: 51.43MIN: 7.96 / MAX: 34.56MIN: 13.67 / MAX: 27.8

PyTorch

Device: CPU - Batch Size: 1 - Model: ResNet-152

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 1 - Model: ResNet-152c7a.2xlargem7i-flex.2xlargec6i.2xlarge510152025SE +/- 0.05, N = 3SE +/- 0.10, N = 3SE +/- 0.01, N = 320.0012.1010.57MIN: 15.6 / MAX: 20.36MIN: 2.89 / MAX: 13.92MIN: 9.04 / MAX: 10.77

PyTorch

Device: CPU - Batch Size: 16 - Model: ResNet-50

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 16 - Model: ResNet-50c7a.2xlargem7i-flex.2xlargec6i.2xlarge714212835SE +/- 0.24, N = 3SE +/- 0.17, N = 3SE +/- 0.12, N = 331.8918.8415.96MIN: 23.54 / MAX: 32.58MIN: 4.37 / MAX: 21.77MIN: 11.65 / MAX: 17.13

PyTorch

Device: CPU - Batch Size: 32 - Model: ResNet-50

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 32 - Model: ResNet-50c7a.2xlargem7i-flex.2xlargec6i.2xlarge714212835SE +/- 0.42, N = 15SE +/- 0.29, N = 15SE +/- 0.17, N = 431.2918.3415.81MIN: 20.73 / MAX: 33.27MIN: 4.11 / MAX: 22.15MIN: 9.11 / MAX: 17.08

PyTorch

Device: CPU - Batch Size: 16 - Model: ResNet-152

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 16 - Model: ResNet-152c7a.2xlargem7i-flex.2xlargec6i.2xlarge3691215SE +/- 0.09, N = 3SE +/- 0.08, N = 5SE +/- 0.05, N = 312.957.366.36MIN: 4.2 / MAX: 13.21MIN: 2.24 / MAX: 8.66MIN: 5.43 / MAX: 6.61

PyTorch

Device: CPU - Batch Size: 32 - Model: ResNet-152

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 32 - Model: ResNet-152c7a.2xlargem7i-flex.2xlargec6i.2xlarge3691215SE +/- 0.05, N = 3SE +/- 0.02, N = 3SE +/- 0.01, N = 313.037.556.38MIN: 10.5 / MAX: 13.2MIN: 2.93 / MAX: 8.62MIN: 3.7 / MAX: 6.57

PyTorch

Device: CPU - Batch Size: 1 - Model: Efficientnet_v2_l

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 1 - Model: Efficientnet_v2_lc7a.2xlargem7i-flex.2xlargec6i.2xlarge3691215SE +/- 0.02, N = 3SE +/- 0.09, N = 12SE +/- 0.02, N = 311.748.997.99MIN: 8.95 / MAX: 11.9MIN: 3.08 / MAX: 10.39MIN: 6.59 / MAX: 8.31

PyTorch

Device: CPU - Batch Size: 16 - Model: Efficientnet_v2_l

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 16 - Model: Efficientnet_v2_lc7a.2xlargem7i-flex.2xlargec6i.2xlarge246810SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.03, N = 38.755.384.06MIN: 5.72 / MAX: 8.84MIN: 2.12 / MAX: 6.08MIN: 3.29 / MAX: 4.32

PyTorch

Device: CPU - Batch Size: 32 - Model: Efficientnet_v2_l

OpenBenchmarking.orgbatches/sec, More Is BetterPyTorch 2.1Device: CPU - Batch Size: 32 - Model: Efficientnet_v2_lc7a.2xlargem7i-flex.2xlargec6i.2xlarge246810SE +/- 0.02, N = 3SE +/- 0.01, N = 3SE +/- 0.02, N = 38.715.474.04MIN: 5.53 / MAX: 8.82MIN: 2.31 / MAX: 6.19MIN: 3.5 / MAX: 4.36

OpenVINO

Model: Face Detection FP16 - Device: CPU

OpenBenchmarking.orgFPS, More Is BetterOpenVINO 2023.2.devModel: Face Detection FP16 - Device: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge246810SE +/- 0.10, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 38.435.161.771. (CXX) g++ options: -fsigned-char -ffunction-sections -fdata-sections -O3 -fno-strict-overflow -fwrapv -pie

OpenVINO

Model: Face Detection FP16-INT8 - Device: CPU

OpenBenchmarking.orgFPS, More Is BetterOpenVINO 2023.2.devModel: Face Detection FP16-INT8 - Device: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge48121620SE +/- 0.16, N = 3SE +/- 0.00, N = 3SE +/- 0.04, N = 316.579.766.531. (CXX) g++ options: -fsigned-char -ffunction-sections -fdata-sections -O3 -fno-strict-overflow -fwrapv -pie

OpenVINO

Model: Machine Translation EN To DE FP16 - Device: CPU

OpenBenchmarking.orgFPS, More Is BetterOpenVINO 2023.2.devModel: Machine Translation EN To DE FP16 - Device: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge1224364860SE +/- 0.02, N = 3SE +/- 0.19, N = 3SE +/- 0.06, N = 354.5153.8722.261. (CXX) g++ options: -fsigned-char -ffunction-sections -fdata-sections -O3 -fno-strict-overflow -fwrapv -pie

Numpy Benchmark

OpenBenchmarking.orgScore, More Is BetterNumpy Benchmarkc7a.2xlargem7i-flex.2xlargec6i.2xlarge130260390520650SE +/- 1.02, N = 3SE +/- 3.53, N = 3SE +/- 1.37, N = 3590.10438.25374.99

PyBench

Total For Average Test Times

OpenBenchmarking.orgMilliseconds, Fewer Is BetterPyBench 2018-02-16Total For Average Test Timesm7i-flex.2xlargec7a.2xlargec6i.2xlarge2004006008001000SE +/- 3.18, N = 3SE +/- 0.67, N = 3SE +/- 0.33, N = 37368871000

oneDNN

Harness: Convolution Batch Shapes Auto - Data Type: f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Convolution Batch Shapes Auto - Data Type: f32 - Engine: CPUm7i-flex.2xlargec6i.2xlargec7a.2xlarge246810SE +/- 0.02435, N = 3SE +/- 0.03862, N = 3SE +/- 0.01338, N = 35.842615.936697.35101MIN: 5MIN: 5.66MIN: 7.221. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Deconvolution Batch shapes_1d - Data Type: f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Deconvolution Batch shapes_1d - Data Type: f32 - Engine: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge3691215SE +/- 0.00229, N = 3SE +/- 0.01302, N = 3SE +/- 0.07954, N = 35.014378.1000712.66180MIN: 4.93MIN: 6.7MIN: 10.81. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Deconvolution Batch shapes_3d - Data Type: f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Deconvolution Batch shapes_3d - Data Type: f32 - Engine: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge246810SE +/- 0.00261, N = 3SE +/- 0.08711, N = 3SE +/- 0.01712, N = 35.103307.948788.09803MIN: 5.05MIN: 6.79MIN: 8.021. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Convolution Batch Shapes Auto - Data Type: u8s8f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Convolution Batch Shapes Auto - Data Type: u8s8f32 - Engine: CPUc6i.2xlargem7i-flex.2xlargec7a.2xlarge246810SE +/- 0.08834, N = 3SE +/- 0.03815, N = 3SE +/- 0.01540, N = 36.243917.194627.74619MIN: 5.78MIN: 6.66MIN: 7.551. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Deconvolution Batch shapes_1d - Data Type: u8s8f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Deconvolution Batch shapes_1d - Data Type: u8s8f32 - Engine: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge0.45860.91721.37581.83442.293SE +/- 0.013619, N = 15SE +/- 0.002971, N = 3SE +/- 0.002826, N = 31.0163271.0942902.038360MIN: 0.79MIN: 1.08MIN: 1.971. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Deconvolution Batch shapes_3d - Data Type: u8s8f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Deconvolution Batch shapes_3d - Data Type: u8s8f32 - Engine: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge0.39890.79781.19671.59561.9945SE +/- 0.011545, N = 4SE +/- 0.004187, N = 3SE +/- 0.013170, N = 31.0032691.2640501.772860MIN: 0.84MIN: 1.25MIN: 1.731. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Recurrent Neural Network Inference - Data Type: f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Recurrent Neural Network Inference - Data Type: f32 - Engine: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge5001000150020002500SE +/- 1.52, N = 3SE +/- 26.45, N = 4SE +/- 7.01, N = 31482.352382.462496.84MIN: 1476.16MIN: 2219.69MIN: 2465.661. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Convolution Batch Shapes Auto - Data Type: bf16bf16bf16 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Convolution Batch Shapes Auto - Data Type: bf16bf16bf16 - Engine: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge816243240SE +/- 0.00275, N = 3SE +/- 0.03984, N = 3SE +/- 0.00337, N = 33.282453.6899333.19200MIN: 33.131. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Deconvolution Batch shapes_1d - Data Type: bf16bf16bf16 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Deconvolution Batch shapes_1d - Data Type: bf16bf16bf16 - Engine: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge1122334455SE +/- 0.00827, N = 3SE +/- 0.00810, N = 3SE +/- 0.01032, N = 31.763506.9020046.69860MIN: 46.421. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Deconvolution Batch shapes_3d - Data Type: bf16bf16bf16 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Deconvolution Batch shapes_3d - Data Type: bf16bf16bf16 - Engine: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge816243240SE +/- 0.01261, N = 3SE +/- 0.00414, N = 3SE +/- 0.00790, N = 33.110753.1233934.71660MIN: 34.61. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Recurrent Neural Network Inference - Data Type: u8s8f32 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Recurrent Neural Network Inference - Data Type: u8s8f32 - Engine: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge5001000150020002500SE +/- 2.09, N = 3SE +/- 24.80, N = 3SE +/- 2.42, N = 31480.932389.642492.29MIN: 1474.24MIN: 2260.99MIN: 2460.061. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

oneDNN

Harness: Recurrent Neural Network Inference - Data Type: bf16bf16bf16 - Engine: CPU

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.3Harness: Recurrent Neural Network Inference - Data Type: bf16bf16bf16 - Engine: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge5001000150020002500SE +/- 1.00, N = 3SE +/- 15.72, N = 3SE +/- 2.60, N = 31478.512318.312501.50MIN: 1472.49MIN: 2205.71MIN: 2476.861. (CXX) g++ options: -O3 -march=native -fopenmp -msse4.1 -fPIC -pie -ldl

OpenVINO

Model: Face Detection FP16 - Device: CPU

OpenBenchmarking.orgms, Fewer Is BetterOpenVINO 2023.2.devModel: Face Detection FP16 - Device: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge5001000150020002500SE +/- 5.97, N = 3SE +/- 0.28, N = 3SE +/- 3.74, N = 3474.46774.232252.04MIN: 427.39 / MAX: 558.54MIN: 767.78 / MAX: 795.49MIN: 2202.18 / MAX: 2317.241. (CXX) g++ options: -fsigned-char -ffunction-sections -fdata-sections -O3 -fno-strict-overflow -fwrapv -pie

OpenVINO

Model: Face Detection FP16-INT8 - Device: CPU

OpenBenchmarking.orgms, Fewer Is BetterOpenVINO 2023.2.devModel: Face Detection FP16-INT8 - Device: CPUm7i-flex.2xlargec7a.2xlargec6i.2xlarge130260390520650SE +/- 2.35, N = 3SE +/- 0.07, N = 3SE +/- 3.28, N = 3241.21409.78610.59MIN: 91.81 / MAX: 374.94MIN: 407.35 / MAX: 424.19MIN: 497.95 / MAX: 643.841. (CXX) g++ options: -fsigned-char -ffunction-sections -fdata-sections -O3 -fno-strict-overflow -fwrapv -pie

OpenVINO

Model: Machine Translation EN To DE FP16 - Device: CPU

OpenBenchmarking.orgms, Fewer Is BetterOpenVINO 2023.2.devModel: Machine Translation EN To DE FP16 - Device: CPUc7a.2xlargem7i-flex.2xlargec6i.2xlarge4080120160200SE +/- 0.02, N = 3SE +/- 0.25, N = 3SE +/- 0.44, N = 373.3474.20179.53MIN: 65.17 / MAX: 80.62MIN: 34.84 / MAX: 96.18MIN: 100.26 / MAX: 3441. (CXX) g++ options: -fsigned-char -ffunction-sections -fdata-sections -O3 -fno-strict-overflow -fwrapv -pie

Phoronix Test Suite v10.8.5