TR 3970X Intel Open Source AMD Ryzen Threadripper 3970X 32-Core testing with a ASUS ROG ZENITH II EXTREME (0702 BIOS) and AMD Radeon VII 16GB on Ubuntu 20.04 via the Phoronix Test Suite.
HTML result view exported from: https://openbenchmarking.org/result/2002127-PTS-TR3970XI48&grr .
TR 3970X Intel Open Source Processor Motherboard Chipset Memory Disk Graphics Audio Monitor Network OS Kernel Desktop Display Server Display Driver OpenGL Compiler File-System Screen Resolution Threadripper 3970X AMD Ryzen Threadripper 3970X 32-Core @ 3.70GHz (32 Cores / 64 Threads) ASUS ROG ZENITH II EXTREME (0702 BIOS) AMD Starship/Matisse 64GB 1000GB Force MP600 AMD Radeon VII 16GB (1801/1000MHz) AMD Vega 20 HDMI Audio ASUS MG28U Aquantia AQC107 NBase-T/IEEE + Intel I211 + Intel Wi-Fi 6 AX200 Ubuntu 20.04 5.6.0-050600rc1-generic (x86_64) 20200209 GNOME Shell 3.34.3 X Server 1.20.7 amdgpu 19.1.0 4.5 Mesa 19.2.4 (LLVM 9.0.0) GCC 9.2.1 20200123 ext4 3840x2160 OpenBenchmarking.org - --build=x86_64-linux-gnu --disable-vtable-verify --disable-werror --enable-bootstrap --enable-checking=release --enable-clocale=gnu --enable-default-pie --enable-gnu-unique-object --enable-languages=c,ada,c++,go,brig,d,fortran,objc,obj-c++,gm2 --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-link-mutex --enable-multiarch --enable-multilib --enable-nls --enable-objc-gc=auto --enable-offload-targets=nvptx-none,hsa --enable-plugin --enable-shared --enable-threads=posix --host=x86_64-linux-gnu --program-prefix=x86_64-linux-gnu- --target=x86_64-linux-gnu --with-abi=m64 --with-arch-32=i686 --with-build-config=bootstrap-lto-lean --with-default-libstdcxx-abi=new --with-gcc-major-version-only --with-multilib-list=m32,m64,mx32 --with-target-system-zlib=auto --with-tune=generic --without-cuda-driver -v - Scaling Governor: acpi-cpufreq ondemand - CPU Microcode: 0x8301025 - itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + spec_store_bypass: Mitigation of SSB disabled via prctl and seccomp + spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization + spectre_v2: Mitigation of Full AMD retpoline IBPB: conditional STIBP: conditional RSB filling + tsx_async_abort: Not affected
TR 3970X Intel Open Source mkl-dnn: Convolution Batch conv_all - f32 mkl-dnn: Convolution Batch conv_all - u8s8f32 mkl-dnn: Deconvolution Batch deconv_all - f32 fftw: Float + SSE - 2D FFT Size 4096 luxcorerender: Rainbow Colors and Prism fftw: Stock - 2D FFT Size 4096 mkl-dnn: Convolution Batch conv_googlenet_v3 - u8s8f32 mkl-dnn: Convolution Batch conv_googlenet_v3 - f32 fftw: Float + SSE - 2D FFT Size 2048 luxcorerender: DLSC mkl-dnn: Convolution Batch conv_3d - u8s8f32 fftw: Stock - 2D FFT Size 2048 mkl-dnn: IP Batch All - f32 mkl-dnn: IP Batch All - u8s8f32 mkl-dnn: Convolution Batch conv_3d - f32 mkl-dnn: Recurrent Neural Network Training - f32 embree: Pathtracer ISPC - Asian Dragon Obj embree: Pathtracer - Asian Dragon Obj fftw: Float + SSE - 2D FFT Size 1024 mkl-dnn: Deconvolution Batch deconv_1d - u8s8f32 mkl-dnn: Deconvolution Batch deconv_1d - f32 fftw: Float + SSE - 1D FFT Size 256 tungsten: Water Caustic mkl-dnn: Convolution Batch conv_alexnet - u8s8f32 embree: Pathtracer ISPC - Crown embree: Pathtracer - Crown mkl-dnn: Convolution Batch conv_alexnet - f32 mkl-dnn: IP Batch 1D - f32 mkl-dnn: IP Batch 1D - u8s8f32 embree: Pathtracer ISPC - Asian Dragon embree: Pathtracer - Asian Dragon mkl-dnn: Deconvolution Batch deconv_3d - u8s8f32 fftw: Stock - 2D FFT Size 1024 tungsten: Hair fftw: Float + SSE - 1D FFT Size 4096 fftw: Float + SSE - 2D FFT Size 512 fftw: Float + SSE - 1D FFT Size 2048 fftw: Stock - 1D FFT Size 4096 fftw: Stock - 2D FFT Size 256 fftw: Float + SSE - 2D FFT Size 256 fftw: Float + SSE - 1D FFT Size 1024 oidn: Memorial fftw: Stock - 1D FFT Size 256 fftw: Float + SSE - 1D FFT Size 512 fftw: Stock - 2D FFT Size 32 fftw: Stock - 2D FFT Size 512 fftw: Float + SSE - 2D FFT Size 32 fftw: Stock - 1D FFT Size 32 fftw: Stock - 1D FFT Size 2048 fftw: Stock - 1D FFT Size 128 tungsten: Volumetric Caustic fftw: Stock - 2D FFT Size 128 mkl-dnn: Deconvolution Batch deconv_3d - f32 fftw: Float + SSE - 2D FFT Size 128 fftw: Stock - 1D FFT Size 1024 fftw: Float + SSE - 1D FFT Size 32 fftw: Stock - 2D FFT Size 64 tungsten: Non-Exponential fftw: Float + SSE - 1D FFT Size 64 fftw: Stock - 1D FFT Size 512 fftw: Stock - 1D FFT Size 64 fftw: Float + SSE - 1D FFT Size 128 fftw: Float + SSE - 2D FFT Size 64 Threadripper 3970X 776.256 12308.9 1270.16 21993 2.41 6858.1 696.080 42.3252 32027 2.99 3724.97 7184.2 11.3691 127.144 4.78292 190.037 36.0048 37.4788 45574 1395.60 1.83211 37548 18.5228 1356.55 36.8651 38.6633 97.9693 2.01809 19.9828 41.7168 41.8483 2327.45 8173.9 8.15037 57225 45047 60512 8800.9 8740.1 44156 58054 24.70 9086.1 50080 11060 8704.0 45811 11273 8989.7 8550.2 3.53144 9178.5 2.70273 48359 9318.3 15489 9556.6 2.66439 20788 9137.0 10428 27900 48212 OpenBenchmarking.org
MKL-DNN DNNL Harness: Convolution Batch conv_all - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_all - Data Type: f32 Threadripper 3970X 200 400 600 800 1000 SE +/- 3.90, N = 3 776.26 MIN: 763.61 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: Convolution Batch conv_all - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_all - Data Type: u8s8f32 Threadripper 3970X 3K 6K 9K 12K 15K SE +/- 16.64, N = 3 12308.9 MIN: 12070.9 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: Deconvolution Batch deconv_all - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_all - Data Type: f32 Threadripper 3970X 300 600 900 1200 1500 SE +/- 2.36, N = 3 1270.16 MIN: 1251.44 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
FFTW Build: Float + SSE - Size: 2D FFT Size 4096 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 4096 Threadripper 3970X 5K 10K 15K 20K 25K SE +/- 7.81, N = 3 21993 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
LuxCoreRender Scene: Rainbow Colors and Prism OpenBenchmarking.org M samples/sec, More Is Better LuxCoreRender 2.2 Scene: Rainbow Colors and Prism Threadripper 3970X 0.5423 1.0846 1.6269 2.1692 2.7115 SE +/- 0.15, N = 15 2.41 MIN: 1.17 / MAX: 3.12
FFTW Build: Stock - Size: 2D FFT Size 4096 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 4096 Threadripper 3970X 1500 3000 4500 6000 7500 SE +/- 8.90, N = 3 6858.1 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
MKL-DNN DNNL Harness: Convolution Batch conv_googlenet_v3 - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_googlenet_v3 - Data Type: u8s8f32 Threadripper 3970X 150 300 450 600 750 SE +/- 1.31, N = 3 696.08 MIN: 676.41 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: Convolution Batch conv_googlenet_v3 - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_googlenet_v3 - Data Type: f32 Threadripper 3970X 10 20 30 40 50 SE +/- 0.27, N = 3 42.33 MIN: 41.2 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
FFTW Build: Float + SSE - Size: 2D FFT Size 2048 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 2048 Threadripper 3970X 7K 14K 21K 28K 35K SE +/- 333.39, N = 3 32027 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
LuxCoreRender Scene: DLSC OpenBenchmarking.org M samples/sec, More Is Better LuxCoreRender 2.2 Scene: DLSC Threadripper 3970X 0.6728 1.3456 2.0184 2.6912 3.364 SE +/- 0.02, N = 3 2.99 MIN: 2.9 / MAX: 3.07
MKL-DNN DNNL Harness: Convolution Batch conv_3d - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_3d - Data Type: u8s8f32 Threadripper 3970X 800 1600 2400 3200 4000 SE +/- 1.09, N = 3 3724.97 MIN: 3712.29 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
FFTW Build: Stock - Size: 2D FFT Size 2048 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 2048 Threadripper 3970X 1500 3000 4500 6000 7500 SE +/- 37.20, N = 3 7184.2 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
MKL-DNN DNNL Harness: IP Batch All - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: IP Batch All - Data Type: f32 Threadripper 3970X 3 6 9 12 15 SE +/- 0.02, N = 3 11.37 MIN: 10.87 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: IP Batch All - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: IP Batch All - Data Type: u8s8f32 Threadripper 3970X 30 60 90 120 150 SE +/- 0.18, N = 3 127.14 MIN: 122.58 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: Convolution Batch conv_3d - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_3d - Data Type: f32 Threadripper 3970X 1.0762 2.1524 3.2286 4.3048 5.381 SE +/- 0.01968, N = 3 4.78292 MIN: 4.59 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: Recurrent Neural Network Training - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Recurrent Neural Network Training - Data Type: f32 Threadripper 3970X 40 80 120 160 200 SE +/- 0.30, N = 3 190.04 MIN: 187.61 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
Embree Binary: Pathtracer ISPC - Model: Asian Dragon Obj OpenBenchmarking.org Frames Per Second, More Is Better Embree 3.6.1 Binary: Pathtracer ISPC - Model: Asian Dragon Obj Threadripper 3970X 8 16 24 32 40 SE +/- 0.01, N = 3 36.00 MIN: 35.74 / MAX: 36.59
Embree Binary: Pathtracer - Model: Asian Dragon Obj OpenBenchmarking.org Frames Per Second, More Is Better Embree 3.6.1 Binary: Pathtracer - Model: Asian Dragon Obj Threadripper 3970X 9 18 27 36 45 SE +/- 0.06, N = 3 37.48 MIN: 37.1 / MAX: 38.17
FFTW Build: Float + SSE - Size: 2D FFT Size 1024 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 1024 Threadripper 3970X 10K 20K 30K 40K 50K SE +/- 116.74, N = 3 45574 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
MKL-DNN DNNL Harness: Deconvolution Batch deconv_1d - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_1d - Data Type: u8s8f32 Threadripper 3970X 300 600 900 1200 1500 SE +/- 4.74, N = 3 1395.60 MIN: 1385.99 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: Deconvolution Batch deconv_1d - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_1d - Data Type: f32 Threadripper 3970X 0.4122 0.8244 1.2366 1.6488 2.061 SE +/- 0.00100, N = 3 1.83211 MIN: 1.78 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
FFTW Build: Float + SSE - Size: 1D FFT Size 256 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 256 Threadripper 3970X 8K 16K 24K 32K 40K SE +/- 528.05, N = 15 37548 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
Tungsten Renderer Scene: Water Caustic OpenBenchmarking.org Seconds, Fewer Is Better Tungsten Renderer 0.2.2 Scene: Water Caustic Threadripper 3970X 5 10 15 20 25 SE +/- 0.01, N = 3 18.52 1. (CXX) g++ options: -std=c++0x -march=znver1 -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4a -mfma -mbmi2 -mno-avx -mno-avx2 -mno-xop -mno-fma4 -mno-avx512f -mno-avx512vl -mno-avx512pf -mno-avx512er -mno-avx512cd -mno-avx512dq -mno-avx512bw -mno-avx512ifma -mno-avx512vbmi -fstrict-aliasing -O3 -rdynamic -ljpeg -lpthread -ldl
MKL-DNN DNNL Harness: Convolution Batch conv_alexnet - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_alexnet - Data Type: u8s8f32 Threadripper 3970X 300 600 900 1200 1500 SE +/- 1.28, N = 3 1356.55 MIN: 1346.13 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
Embree Binary: Pathtracer ISPC - Model: Crown OpenBenchmarking.org Frames Per Second, More Is Better Embree 3.6.1 Binary: Pathtracer ISPC - Model: Crown Threadripper 3970X 8 16 24 32 40 SE +/- 0.06, N = 3 36.87 MIN: 36.37 / MAX: 37.82
Embree Binary: Pathtracer - Model: Crown OpenBenchmarking.org Frames Per Second, More Is Better Embree 3.6.1 Binary: Pathtracer - Model: Crown Threadripper 3970X 9 18 27 36 45 SE +/- 0.10, N = 3 38.66 MIN: 38.14 / MAX: 39.58
MKL-DNN DNNL Harness: Convolution Batch conv_alexnet - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_alexnet - Data Type: f32 Threadripper 3970X 20 40 60 80 100 SE +/- 0.30, N = 3 97.97 MIN: 96.76 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: IP Batch 1D - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: IP Batch 1D - Data Type: f32 Threadripper 3970X 0.4541 0.9082 1.3623 1.8164 2.2705 SE +/- 0.00613, N = 3 2.01809 MIN: 1.95 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
MKL-DNN DNNL Harness: IP Batch 1D - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: IP Batch 1D - Data Type: u8s8f32 Threadripper 3970X 5 10 15 20 25 SE +/- 0.05, N = 3 19.98 MIN: 19.06 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
Embree Binary: Pathtracer ISPC - Model: Asian Dragon OpenBenchmarking.org Frames Per Second, More Is Better Embree 3.6.1 Binary: Pathtracer ISPC - Model: Asian Dragon Threadripper 3970X 10 20 30 40 50 SE +/- 0.03, N = 3 41.72 MIN: 41.42 / MAX: 42.36
Embree Binary: Pathtracer - Model: Asian Dragon OpenBenchmarking.org Frames Per Second, More Is Better Embree 3.6.1 Binary: Pathtracer - Model: Asian Dragon Threadripper 3970X 10 20 30 40 50 SE +/- 0.06, N = 3 41.85 MIN: 41.47 / MAX: 42.6
MKL-DNN DNNL Harness: Deconvolution Batch deconv_3d - Data Type: u8s8f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_3d - Data Type: u8s8f32 Threadripper 3970X 500 1000 1500 2000 2500 SE +/- 12.87, N = 3 2327.45 MIN: 2305.28 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
FFTW Build: Stock - Size: 2D FFT Size 1024 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 1024 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 12.92, N = 3 8173.9 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
Tungsten Renderer Scene: Hair OpenBenchmarking.org Seconds, Fewer Is Better Tungsten Renderer 0.2.2 Scene: Hair Threadripper 3970X 2 4 6 8 10 SE +/- 0.00414, N = 3 8.15037 1. (CXX) g++ options: -std=c++0x -march=znver1 -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4a -mfma -mbmi2 -mno-avx -mno-avx2 -mno-xop -mno-fma4 -mno-avx512f -mno-avx512vl -mno-avx512pf -mno-avx512er -mno-avx512cd -mno-avx512dq -mno-avx512bw -mno-avx512ifma -mno-avx512vbmi -fstrict-aliasing -O3 -rdynamic -ljpeg -lpthread -ldl
FFTW Build: Float + SSE - Size: 1D FFT Size 4096 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 4096 Threadripper 3970X 12K 24K 36K 48K 60K SE +/- 549.89, N = 3 57225 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 2D FFT Size 512 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 512 Threadripper 3970X 10K 20K 30K 40K 50K SE +/- 87.88, N = 3 45047 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 1D FFT Size 2048 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 2048 Threadripper 3970X 13K 26K 39K 52K 65K SE +/- 548.16, N = 3 60512 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 4096 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 4096 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 5.26, N = 3 8800.9 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 2D FFT Size 256 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 256 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 76.80, N = 3 8740.1 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 2D FFT Size 256 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 256 Threadripper 3970X 9K 18K 27K 36K 45K SE +/- 508.53, N = 3 44156 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 1D FFT Size 1024 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 1024 Threadripper 3970X 12K 24K 36K 48K 60K SE +/- 302.03, N = 3 58054 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
Intel Open Image Denoise Scene: Memorial OpenBenchmarking.org Images / Sec, More Is Better Intel Open Image Denoise 1.0.0 Scene: Memorial Threadripper 3970X 6 12 18 24 30 SE +/- 0.03, N = 3 24.70
FFTW Build: Stock - Size: 1D FFT Size 256 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 256 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 24.48, N = 3 9086.1 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 1D FFT Size 512 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 512 Threadripper 3970X 11K 22K 33K 44K 55K SE +/- 78.30, N = 3 50080 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 2D FFT Size 32 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 32 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 12.12, N = 3 11060 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 2D FFT Size 512 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 512 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 89.15, N = 3 8704.0 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 2D FFT Size 32 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 32 Threadripper 3970X 10K 20K 30K 40K 50K SE +/- 112.90, N = 3 45811 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 32 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 32 Threadripper 3970X 2K 4K 6K 8K 10K 11273 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 2048 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 2048 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 5.17, N = 3 8989.7 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 128 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 128 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 9.65, N = 3 8550.2 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
Tungsten Renderer Scene: Volumetric Caustic OpenBenchmarking.org Seconds, Fewer Is Better Tungsten Renderer 0.2.2 Scene: Volumetric Caustic Threadripper 3970X 0.7946 1.5892 2.3838 3.1784 3.973 SE +/- 0.00240, N = 3 3.53144 1. (CXX) g++ options: -std=c++0x -march=znver1 -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4a -mfma -mbmi2 -mno-avx -mno-avx2 -mno-xop -mno-fma4 -mno-avx512f -mno-avx512vl -mno-avx512pf -mno-avx512er -mno-avx512cd -mno-avx512dq -mno-avx512bw -mno-avx512ifma -mno-avx512vbmi -fstrict-aliasing -O3 -rdynamic -ljpeg -lpthread -ldl
FFTW Build: Stock - Size: 2D FFT Size 128 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 128 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 11.46, N = 3 9178.5 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
MKL-DNN DNNL Harness: Deconvolution Batch deconv_3d - Data Type: f32 OpenBenchmarking.org ms, Fewer Is Better MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_3d - Data Type: f32 Threadripper 3970X 0.6081 1.2162 1.8243 2.4324 3.0405 SE +/- 0.00427, N = 3 2.70273 MIN: 2.63 1. (CXX) g++ options: -O3 -march=native -std=c++11 -msse4.1 -fPIC -fopenmp -pie -lpthread -ldl
FFTW Build: Float + SSE - Size: 2D FFT Size 128 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 128 Threadripper 3970X 10K 20K 30K 40K 50K SE +/- 437.77, N = 3 48359 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 1024 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 1024 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 24.99, N = 3 9318.3 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 1D FFT Size 32 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 32 Threadripper 3970X 3K 6K 9K 12K 15K SE +/- 31.01, N = 3 15489 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 2D FFT Size 64 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 2D FFT Size 64 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 130.25, N = 3 9556.6 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
Tungsten Renderer Scene: Non-Exponential OpenBenchmarking.org Seconds, Fewer Is Better Tungsten Renderer 0.2.2 Scene: Non-Exponential Threadripper 3970X 0.5995 1.199 1.7985 2.398 2.9975 SE +/- 0.01210, N = 3 2.66439 1. (CXX) g++ options: -std=c++0x -march=znver1 -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4a -mfma -mbmi2 -mno-avx -mno-avx2 -mno-xop -mno-fma4 -mno-avx512f -mno-avx512vl -mno-avx512pf -mno-avx512er -mno-avx512cd -mno-avx512dq -mno-avx512bw -mno-avx512ifma -mno-avx512vbmi -fstrict-aliasing -O3 -rdynamic -ljpeg -lpthread -ldl
FFTW Build: Float + SSE - Size: 1D FFT Size 64 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 64 Threadripper 3970X 4K 8K 12K 16K 20K SE +/- 75.35, N = 3 20788 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 512 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 512 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 18.81, N = 3 9137.0 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Stock - Size: 1D FFT Size 64 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Stock - Size: 1D FFT Size 64 Threadripper 3970X 2K 4K 6K 8K 10K SE +/- 2.73, N = 3 10428 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 1D FFT Size 128 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 1D FFT Size 128 Threadripper 3970X 6K 12K 18K 24K 30K SE +/- 226.32, N = 3 27900 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
FFTW Build: Float + SSE - Size: 2D FFT Size 64 OpenBenchmarking.org Mflops, More Is Better FFTW 3.3.6 Build: Float + SSE - Size: 2D FFT Size 64 Threadripper 3970X 10K 20K 30K 40K 50K SE +/- 200.21, N = 3 48212 1. (CC) gcc options: -pthread -O3 -fomit-frame-pointer -mtune=native -malign-double -fstrict-aliasing -fno-schedule-insns -ffast-math -lm
Phoronix Test Suite v10.8.5