GCC 13 Compiler Benchmarks AMD EPYC Genoa

AMD EPYC 9654 GCC 13 development compiler benchmarks by Michael Larabel for a future article.

Compare your own system(s) to this result file with the Phoronix Test Suite by running the command: phoronix-test-suite benchmark 2301045-NE-GCC13DEVE69
Jump To Table - Results

View

Do Not Show Noisy Results
Do Not Show Results With Incomplete Data
Do Not Show Results With Little Change/Spread
List Notable Results
Show Result Confidence Charts

Limit displaying results to tests within:

AV1 2 Tests
C/C++ Compiler Tests 8 Tests
CPU Massive 7 Tests
Creator Workloads 11 Tests
Cryptography 4 Tests
Encoding 4 Tests
HPC - High Performance Computing 8 Tests
Imaging 5 Tests
Molecular Dynamics 3 Tests
MPI Benchmarks 3 Tests
Multi-Core 11 Tests
OpenMPI Tests 4 Tests
Programmer / Developer System Benchmarks 3 Tests
Python Tests 2 Tests
Scientific Computing 6 Tests
Server 2 Tests
Server CPU Tests 6 Tests
Video Encoding 3 Tests

Statistics

Show Overall Harmonic Mean(s)
Show Overall Geometric Mean
Show Geometric Means Per-Suite/Category
Show Wins / Losses Counts (Pie Chart)
Normalize Results
Remove Outliers Before Calculating Averages

Graph Settings

Force Line Graphs Where Applicable
Convert To Scalar Where Applicable
Prefer Vertical Bar Graphs

Multi-Way Comparison

Condense Multi-Option Tests Into Single Result Graphs

Table

Show Detailed System Result Table

Run Management

Highlight
Result
Hide
Result
Result
Identifier
Performance Per
Dollar
Date
Run
  Test
  Duration
Znver4
January 03 2023
  5 Hours, 43 Minutes
Znver4 + Prefer AVX-512
January 03 2023
  5 Hours, 30 Minutes
Znver3
January 04 2023
  6 Hours, 7 Minutes
Znver3 + AVX-512
January 04 2023
  6 Hours, 17 Minutes
Invert Hiding All Results Option
  5 Hours, 54 Minutes

Only show results where is faster than
Only show results matching title/arguments (delimit multiple options with a comma):
Do not show results matching title/arguments (delimit multiple options with a comma):

Znver4

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=native -flto" CFLAGS="-O3 -march=native -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Znver4 + Prefer AVX-512

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=native -mprefer-vector-width=512 -flto" CFLAGS="-O3 -march=native -mprefer-vector-width=512 -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Znver3

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=znver3 -flto" CFLAGS="-O3 -march=znver3 -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Znver3 + AVX-512

Processor: 2 x AMD EPYC 9654 96-Core @ 3.71GHz (192 Cores / 384 Threads), Motherboard: AMD Titanite_4G (RTI1002E BIOS), Chipset: AMD Device 14a4, Memory: 1520GB, Disk: 800GB INTEL SSDPF21Q800GB, Graphics: ASPEED, Monitor: VGA HDMI, Network: Broadcom NetXtreme BCM5720 PCIe

OS: Ubuntu 23.04, Kernel: 5.19.0-21-generic (x86_64), Desktop: GNOME Shell 43.1, Display Server: X Server 1.21.1.4, Vulkan: 1.3.224, Compiler: GCC 13.0.0 20230103, File-System: ext4, Screen Resolution: 1920x1080

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -flto" CFLAGS="-O3 -march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

GCC 13 Compiler Benchmarks AMD EPYC GenoaOpenBenchmarking.orgPhoronix Test Suite2 x AMD EPYC 9654 96-Core @ 3.71GHz (192 Cores / 384 Threads)AMD Titanite_4G (RTI1002E BIOS)AMD Device 14a41520GB800GB INTEL SSDPF21Q800GBASPEEDVGA HDMIBroadcom NetXtreme BCM5720 PCIeUbuntu 23.045.19.0-21-generic (x86_64)GNOME Shell 43.1X Server 1.21.1.41.3.224GCC 13.0.0 20230103ext41920x1080ProcessorMotherboardChipsetMemoryDiskGraphicsMonitorNetworkOSKernelDesktopDisplay ServerVulkanCompilerFile-SystemScreen ResolutionGCC 13 Compiler Benchmarks AMD EPYC Genoa PerformanceSystem Logs- Transparent Huge Pages: madvise- Znver4: CXXFLAGS="-O3 -march=native -flto" CFLAGS="-O3 -march=native -flto"- Znver4 + Prefer AVX-512: CXXFLAGS="-O3 -march=native -mprefer-vector-width=512 -flto" CFLAGS="-O3 -march=native -mprefer-vector-width=512 -flto" - Znver3: CXXFLAGS="-O3 -march=znver3 -flto" CFLAGS="-O3 -march=znver3 -flto"- Znver3 + AVX-512: CXXFLAGS="-O3 -march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -flto" CFLAGS="-O3 -march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -flto" - --disable-multilib --enable-checking=release- Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d - Python 3.10.9- itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Znver4Znver4 + Prefer AVX-512Znver3Znver3 + AVX-512Result OverviewPhoronix Test Suite100%109%118%127%Cpuminer-OptCP2K Molecular DynamicsGROMACSKripkeGraphicsMagickASTC EncoderGPAWSVT-AV1WebP Image EncodeJPEG XL Decoding libjxlCoremarkStargate Digital Audio Workstationlibavif avifencZstd CompressionKvazaarJPEG XL libjxlsimdjsonNgspiceminiBUDELiquid-DSPSecureMarkoneDNNLAMMPS Molecular Dynamics SimulatorQuantLibPJSIPACES DGEMMOpenSSLSMHasher

GCC 13 Compiler Benchmarks AMD EPYC Genoajpegxl: JPEG - 100cp2k: Fayalite-FISTonednn: Recurrent Neural Network Inference - u8s8f32 - CPUgraphics-magick: Resizingjpegxl: PNG - 100securemark: SecureMark-TLSonednn: Recurrent Neural Network Training - bf16bf16bf16 - CPUgraphics-magick: Noise-Gaussianonednn: Recurrent Neural Network Training - u8s8f32 - CPUcompress-zstd: 19, Long Mode - Decompression Speedcompress-zstd: 19, Long Mode - Compression Speedgraphics-magick: HWB Color Spaceopenssl: SHA256onednn: Recurrent Neural Network Inference - bf16bf16bf16 - CPUlammps: 20k Atomsjpegxl: JPEG - 90jpegxl: PNG - 90ngspice: C2670ngspice: C7552compress-zstd: 19 - Decompression Speedcompress-zstd: 19 - Compression Speedgraphics-magick: Swirlpjsip: INVITEonednn: IP Shapes 1D - bf16bf16bf16 - CPUpjsip: OPTIONS, Statefulavifenc: 0stargate: 192000 - 1024simdjson: DistinctUserIDsimdjson: PartialTweetssimdjson: TopTweetopenssl: RSA4096openssl: RSA4096graphics-magick: Sharpengraphics-magick: Rotategraphics-magick: Enhancedsimdjson: Kostyajpegxl-decode: 1onednn: IP Shapes 1D - u8s8f32 - CPUsimdjson: LargeRandkripke: webp: Quality 100, Lossless, Highest Compressiononednn: IP Shapes 3D - bf16bf16bf16 - CPUstargate: 96000 - 1024cpuminer-opt: Deepcoinsvt-av1: Preset 4 - Bosphorus 4Kavifenc: 2cpuminer-opt: Quad SHA-256, Pyritepjsip: OPTIONS, Statelesscpuminer-opt: Garlicoincpuminer-opt: LBC, LBRY Creditscpuminer-opt: scryptcpuminer-opt: Triple SHA-256, Onecoincpuminer-opt: x25xcpuminer-opt: Skeincoincpuminer-opt: Magiquantlib: onednn: Deconvolution Batch shapes_1d - u8s8f32 - CPUgpaw: Carbon Nanotubesvt-av1: Preset 12 - Bosphorus 4Kjpegxl-decode: Allkvazaar: Bosphorus 4K - Ultra Fastgromacs: MPI CPU - water_GMX50_barecoremark: CoreMark Size 666 - Iterations Per Secondonednn: Deconvolution Batch shapes_1d - bf16bf16bf16 - CPUsvt-av1: Preset 13 - Bosphorus 4Kliquid-dsp: 384 - 256 - 57liquid-dsp: 256 - 256 - 57liquid-dsp: 128 - 256 - 57minibude: OpenMP - BM2minibude: OpenMP - BM2webp: Quality 100, Losslesssmhasher: MeowHash x86_64 AES-NIsmhasher: MeowHash x86_64 AES-NIonednn: Convolution Batch Shapes Auto - bf16bf16bf16 - CPUonednn: Matrix Multiply Batch Shapes Transformer - bf16bf16bf16 - CPUsvt-av1: Preset 8 - Bosphorus 4Kkvazaar: Bosphorus 4K - Mediumkvazaar: Bosphorus 4K - Very Fastonednn: IP Shapes 3D - u8s8f32 - CPUsmhasher: FarmHash32 x86_64 AVXsmhasher: FarmHash32 x86_64 AVXonednn: Deconvolution Batch shapes_3d - u8s8f32 - CPUastcenc: Mediumastcenc: Exhaustivewebp: Quality 100, Highest Compressionsmhasher: t1ha0_aes_avx2 x86_64smhasher: t1ha0_aes_avx2 x86_64onednn: Convolution Batch Shapes Auto - u8s8f32 - CPUastcenc: Thoroughavifenc: 6, Losslessminibude: OpenMP - BM1minibude: OpenMP - BM1avifenc: 10, Losslessmt-dgemm: Sustained Floating-Point Rateonednn: Deconvolution Batch shapes_3d - bf16bf16bf16 - CPUlammps: Rhodopsin Proteinavifenc: 6webp: Quality 100webp: DefaultZnver4Znver4 + Prefer AVX-512Znver3Znver3 + AVX-5120.731174.6652405.63880.832965482134.8510242020.243708.840.811802653267135872479.7255.6189.489.8395.06692.1023584.1102.12862520021.1359923761.1222.7837176.536.636.962939503.544490.1135967322344.1748.6414.81681.272615622800.574.266004.3312301591475.39234.10422510673357677241310654874790.1133066438042.8820147708440.793096.90.93716422.350210.386277.2078.6919.4937694653.8990162.33363196.1151124933333397895000006990866667264.1476603.6841.4544.97754272.140.4438910.50057993.75664.3570.710.88664226.49440565.070.274769493.224812.94123.2520.524102354.570.380486118.73924.398214.5025362.5423.54170.0529520.65810151.5852.34711.5418.950.751263.7072359.29870.822941222070.1310132123.353685.742.511672662301240702444.5255.7399.439.8195.59993.4533581.6105.32681508423.5341928861.0732.8672336.866.436.912924586.444301.8131465621504.1948.2114.29281.242546485330.584.333664.4082901622425.42333.90223239953367917213010858274782.7433012178217.7020093678467.243112.60.9541422.819196.804272.3174.9819.4377861097.8566402.29880208.6431127066666798098000006977633333265.5206638.0071.4744.95054297.040.4455320.48397994.34064.0570.810.88130326.50340563.960.274662420.699213.09363.2320.806102399.740.388483118.99324.462211.0175275.4263.57270.2982820.65006752.2872.31711.5418.970.741211.1262442.35890.822940572093.7510182108.463695.539.811342664643611932418.2855.7219.189.6195.17592.9433574.9104.42563514919.5885922661.6582.8200796.466.746.612938488.544435.3128564518374.2048.2214.61791.252638128470.584.524404.3564461649935.36033.9111378987336885495234970202959.1532552536116.9714140478490.733120.50.94701321.721222.902269.5978.0118.2317640546.2720742.31858219.5701117600000097356666676940833333264.2846607.0911.4744.95754281.940.4427480.51093995.11664.2772.010.90794126.49440565.720.279655459.692412.22213.6420.533102351.870.392765116.09124.437214.0785351.9443.58170.3780650.65634251.4552.33111.4818.990.711407.1252377.17860.812965752103.419752099.433684.743.910622668990894532531.4256.0369.279.8695.18693.5243594.8102.92826513222.9431923662.6842.8683736.616.686.732935372.144499.3132160522084.1647.0214.250211.252717357080.584.336584.4133791601575.37434.23022647473366157283710677434763.9133236807941.3820049908355.753114.90.93637422.823206.676266.5376.7619.0877871273.9327622.36166210.4261130166666798137000006999233333266.1556653.8641.4544.96854284.650.4312990.50484292.96463.8070.710.90230326.49440559.330.276965511.524813.03053.1120.527102403.920.392875117.69104.514214.6255365.6123.64770.1864670.64708751.8162.40611.3818.85OpenBenchmarking.org

JPEG XL libjxl

The JPEG XL Image Coding System is designed to provide next-generation JPEG image capabilities with JPEG XL offering better image quality and compression over legacy JPEG. This test profile is currently focused on the multi-threaded JPEG XL image encode performance using the reference libjxl library. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterJPEG XL libjxl 0.7Input: JPEG - Quality: 100Znver3 + AVX-512Znver4Znver3Znver4 + Prefer AVX-5120.16880.33760.50640.67520.844SE +/- 0.01, N = 9SE +/- 0.01, N = 9SE +/- 0.01, N = 6SE +/- 0.01, N = 30.710.730.740.75-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CXX) g++ options: -O3 -flto -fno-rtti -funwind-tables -O2 -fPIE -pie -latomic

CP2K Molecular Dynamics

CP2K is an open-source molecular dynamics software package focused on quantum chemistry and solid-state physics. This test profile currently uses the SSMP (OpenMP) version of cp2k. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is BetterCP2K Molecular Dynamics 8.2Input: Fayalite-FISTZnver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver4300600900120015001407.131263.711211.131174.67

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Recurrent Neural Network Inference - Data Type: u8s8f32 - Engine: CPUZnver3Znver4Znver3 + AVX-512Znver4 + Prefer AVX-5125001000150020002500SE +/- 32.77, N = 12SE +/- 20.65, N = 15SE +/- 26.39, N = 15SE +/- 31.29, N = 152442.352405.632377.172359.29-march=znver3 - MIN: 2123.49MIN: 2158.4-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 2080.2MIN: 2066.841. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

GraphicsMagick

This is a test of GraphicsMagick with its OpenMP implementation that performs various imaging tests on a sample 6000x4000 pixel JPEG image. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: ResizingZnver3 + AVX-512Znver4 + Prefer AVX-512Znver4Znver320406080100SE +/- 0.90, N = 15SE +/- 0.94, N = 15SE +/- 1.00, N = 15SE +/- 1.27, N = 1586878889-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native-march=znver31. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

JPEG XL libjxl

The JPEG XL Image Coding System is designed to provide next-generation JPEG image capabilities with JPEG XL offering better image quality and compression over legacy JPEG. This test profile is currently focused on the multi-threaded JPEG XL image encode performance using the reference libjxl library. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterJPEG XL libjxl 0.7Input: PNG - Quality: 100Znver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver40.18680.37360.56040.74720.934SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 30.810.820.820.83-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CXX) g++ options: -O3 -flto -fno-rtti -funwind-tables -O2 -fPIE -pie -latomic

SecureMark

SecureMark is an objective, standardized benchmarking framework for measuring the efficiency of cryptographic processing solutions developed by EEMBC. SecureMark-TLS is benchmarking Transport Layer Security performance with a focus on IoT/edge computing. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgmarks, More Is BetterSecureMark 1.0.4Benchmark: SecureMark-TLSZnver3Znver4 + Prefer AVX-512Znver4Znver3 + AVX-51260K120K180K240K300KSE +/- 1276.92, N = 3SE +/- 640.61, N = 3SE +/- 383.28, N = 3SE +/- 380.50, N = 32940572941222965482965751. (CC) gcc options: -pedantic -O3

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Recurrent Neural Network Training - Data Type: bf16bf16bf16 - Engine: CPUZnver4Znver3 + AVX-512Znver3Znver4 + Prefer AVX-5125001000150020002500SE +/- 9.82, N = 3SE +/- 15.62, N = 15SE +/- 27.68, N = 15SE +/- 16.85, N = 32134.852103.412093.752070.13MIN: 2008.7-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 1883.21-march=znver3 - MIN: 1826.26MIN: 1938.321. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

GraphicsMagick

This is a test of GraphicsMagick with its OpenMP implementation that performs various imaging tests on a sample 6000x4000 pixel JPEG image. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: Noise-GaussianZnver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver42004006008001000SE +/- 11.43, N = 15SE +/- 11.39, N = 15SE +/- 5.13, N = 3SE +/- 6.60, N = 15975101310181024-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Recurrent Neural Network Training - Data Type: u8s8f32 - Engine: CPUZnver4 + Prefer AVX-512Znver3Znver3 + AVX-512Znver45001000150020002500SE +/- 26.64, N = 12SE +/- 26.49, N = 3SE +/- 18.84, N = 15SE +/- 16.34, N = 32123.352108.462099.432020.24MIN: 1873.09-march=znver3 - MIN: 1945.83-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 1917.53MIN: 1878.541. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

Zstd Compression

This test measures the time needed to compress/decompress a sample file (a FreeBSD disk image - FreeBSD-12.2-RELEASE-amd64-memstick.img) using Zstd compression with options for different compression levels / settings. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMB/s, More Is BetterZstd Compression 1.5.0Compression Level: 19, Long Mode - Decompression SpeedZnver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver48001600240032004000SE +/- 12.89, N = 15SE +/- 13.90, N = 15SE +/- 11.99, N = 15SE +/- 12.59, N = 153684.73685.73695.53708.8-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CC) gcc options: -O3 -flto -pthread -lz -llzma

OpenBenchmarking.orgMB/s, More Is BetterZstd Compression 1.5.0Compression Level: 19, Long Mode - Compression SpeedZnver3Znver4Znver4 + Prefer AVX-512Znver3 + AVX-5121020304050SE +/- 0.40, N = 15SE +/- 0.40, N = 15SE +/- 0.59, N = 15SE +/- 0.79, N = 1539.840.842.543.9-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -O3 -flto -pthread -lz -llzma

GraphicsMagick

This is a test of GraphicsMagick with its OpenMP implementation that performs various imaging tests on a sample 6000x4000 pixel JPEG image. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: HWB Color SpaceZnver3 + AVX-512Znver3Znver4 + Prefer AVX-512Znver430060090012001500SE +/- 1.86, N = 3SE +/- 11.42, N = 15SE +/- 15.31, N = 15SE +/- 4.84, N = 31062113411671180-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native-march=native1. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

OpenSSL

OpenSSL is an open-source toolkit that implements SSL (Secure Sockets Layer) and TLS (Transport Layer Security) protocols. This test profile makes use of the built-in "openssl speed" benchmarking capabilities. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgbyte/s, More Is BetterOpenSSL 3.0Algorithm: SHA256Znver4Znver4 + Prefer AVX-512Znver3Znver3 + AVX-51260000M120000M180000M240000M300000MSE +/- 150345864.28, N = 3SE +/- 172236238.05, N = 3SE +/- 18811489.95, N = 3SE +/- 143443542.20, N = 3265326713587266230124070266464361193266899089453-march=native-march=native-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -pthread -m64 -O3 -flto -lssl -lcrypto -ldl

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Recurrent Neural Network Inference - Data Type: bf16bf16bf16 - Engine: CPUZnver3 + AVX-512Znver4Znver4 + Prefer AVX-512Znver35001000150020002500SE +/- 16.53, N = 3SE +/- 24.81, N = 3SE +/- 25.52, N = 3SE +/- 24.11, N = 152531.422479.722444.522418.28-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 2281.87MIN: 2315.89MIN: 2258.45-march=znver3 - MIN: 2149.671. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

LAMMPS Molecular Dynamics Simulator

LAMMPS is a classical molecular dynamics code, and an acronym for Large-scale Atomic/Molecular Massively Parallel Simulator. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgns/day, More Is BetterLAMMPS Molecular Dynamics Simulator 23Jun2022Model: 20k AtomsZnver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-5121326395265SE +/- 0.15, N = 3SE +/- 0.04, N = 3SE +/- 0.21, N = 3SE +/- 0.08, N = 355.6255.7255.7456.04-march=native-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -lm -ldl

JPEG XL libjxl

The JPEG XL Image Coding System is designed to provide next-generation JPEG image capabilities with JPEG XL offering better image quality and compression over legacy JPEG. This test profile is currently focused on the multi-threaded JPEG XL image encode performance using the reference libjxl library. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterJPEG XL libjxl 0.7Input: JPEG - Quality: 90Znver3Znver3 + AVX-512Znver4 + Prefer AVX-512Znver43691215SE +/- 0.03, N = 3SE +/- 0.13, N = 3SE +/- 0.03, N = 3SE +/- 0.05, N = 39.189.279.439.48-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto -fno-rtti -funwind-tables -O2 -fPIE -pie -latomic

OpenBenchmarking.orgMP/s, More Is BetterJPEG XL libjxl 0.7Input: PNG - Quality: 90Znver3Znver4 + Prefer AVX-512Znver4Znver3 + AVX-5123691215SE +/- 0.01, N = 3SE +/- 0.04, N = 3SE +/- 0.03, N = 3SE +/- 0.07, N = 39.619.819.839.86-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -fno-rtti -funwind-tables -O2 -fPIE -pie -latomic

Ngspice

Ngspice is an open-source SPICE circuit simulator. Ngspice was originally based on the Berkeley SPICE electronic circuit simulator. Ngspice supports basic threading using OpenMP. This test profile is making use of the ISCAS 85 benchmark circuits. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is BetterNgspice 34Circuit: C2670Znver4 + Prefer AVX-512Znver3 + AVX-512Znver3Znver420406080100SE +/- 0.17, N = 3SE +/- 0.24, N = 3SE +/- 0.09, N = 3SE +/- 0.76, N = 395.6095.1995.1895.07-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native1. (CC) gcc options: -O3 -flto -fopenmp -lm -lfftw3 -lXaw -lXmu -lXt -lXext -lX11 -lSM -lICE

OpenBenchmarking.orgSeconds, Fewer Is BetterNgspice 34Circuit: C7552Znver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver420406080100SE +/- 0.28, N = 3SE +/- 0.14, N = 3SE +/- 0.28, N = 3SE +/- 1.00, N = 393.5293.4592.9492.10-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CC) gcc options: -O3 -flto -fopenmp -lm -lfftw3 -lXaw -lXmu -lXt -lXext -lX11 -lSM -lICE

Zstd Compression

This test measures the time needed to compress/decompress a sample file (a FreeBSD disk image - FreeBSD-12.2-RELEASE-amd64-memstick.img) using Zstd compression with options for different compression levels / settings. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMB/s, More Is BetterZstd Compression 1.5.0Compression Level: 19 - Decompression SpeedZnver3Znver4 + Prefer AVX-512Znver4Znver3 + AVX-5128001600240032004000SE +/- 12.56, N = 15SE +/- 36.60, N = 3SE +/- 17.00, N = 6SE +/- 32.49, N = 33574.93581.63584.13594.8-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -O3 -flto -pthread -lz -llzma

OpenBenchmarking.orgMB/s, More Is BetterZstd Compression 1.5.0Compression Level: 19 - Compression SpeedZnver4Znver3 + AVX-512Znver3Znver4 + Prefer AVX-51220406080100SE +/- 1.03, N = 6SE +/- 1.27, N = 3SE +/- 0.91, N = 15SE +/- 1.30, N = 3102.1102.9104.4105.3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native1. (CC) gcc options: -O3 -flto -pthread -lz -llzma

GraphicsMagick

This is a test of GraphicsMagick with its OpenMP implementation that performs various imaging tests on a sample 6000x4000 pixel JPEG image. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: SwirlZnver3Znver4 + Prefer AVX-512Znver3 + AVX-512Znver46001200180024003000SE +/- 36.67, N = 3SE +/- 8.09, N = 3SE +/- 28.99, N = 3SE +/- 26.19, N = 72563268128262862-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

PJSIP

PJSIP is a free and open source multimedia communication library written in C language implementing standard based protocols such as SIP, SDP, RTP, STUN, TURN, and ICE. It combines signaling protocol (SIP) with rich multimedia framework and NAT traversal functionality into high level API that is portable and suitable for almost any type of systems ranging from desktops, embedded systems, to mobile handsets. This test profile is making use of pjsip-perf with both the client/server on teh system. More details on the PJSIP benchmark at https://www.pjsip.org/high-performance-sip.htm Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgResponses Per Second, More Is BetterPJSIP 2.11Method: INVITEZnver4 + Prefer AVX-512Znver3 + AVX-512Znver3Znver411002200330044005500SE +/- 51.36, N = 5SE +/- 24.85, N = 3SE +/- 15.00, N = 3SE +/- 19.91, N = 35084513251495200-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native1. (CC) gcc options: -lavformat -lavcodec -lswscale -lavutil -lstdc++ -lopus -lssl -lcrypto -lm -lrt -lpthread -lasound -O3 -flto

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: IP Shapes 1D - Data Type: bf16bf16bf16 - Engine: CPUZnver4 + Prefer AVX-512Znver3 + AVX-512Znver4Znver3612182430SE +/- 0.56, N = 12SE +/- 0.76, N = 15SE +/- 0.95, N = 15SE +/- 0.82, N = 1523.5322.9421.1419.59MIN: 10.06-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 9.35MIN: 9.61-march=znver3 - MIN: 9.211. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

PJSIP

PJSIP is a free and open source multimedia communication library written in C language implementing standard based protocols such as SIP, SDP, RTP, STUN, TURN, and ICE. It combines signaling protocol (SIP) with rich multimedia framework and NAT traversal functionality into high level API that is portable and suitable for almost any type of systems ranging from desktops, embedded systems, to mobile handsets. This test profile is making use of pjsip-perf with both the client/server on teh system. More details on the PJSIP benchmark at https://www.pjsip.org/high-performance-sip.htm Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgResponses Per Second, More Is BetterPJSIP 2.11Method: OPTIONS, StatefulZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-5122K4K6K8K10KSE +/- 72.95, N = 3SE +/- 36.35, N = 3SE +/- 23.25, N = 3SE +/- 32.42, N = 39226923692379288-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CC) gcc options: -lavformat -lavcodec -lswscale -lavutil -lstdc++ -lopus -lssl -lcrypto -lm -lrt -lpthread -lasound -O3 -flto

libavif avifenc

This is a test of the AOMedia libavif library testing the encoding of a JPEG image to AV1 Image Format (AVIF). Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is Betterlibavif avifenc 0.11Encoder Speed: 0Znver3 + AVX-512Znver3Znver4Znver4 + Prefer AVX-5121428425670SE +/- 0.02, N = 3SE +/- 0.30, N = 3SE +/- 0.13, N = 3SE +/- 0.07, N = 362.6861.6661.1261.07-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native-march=native1. (CXX) g++ options: -O3 -fPIC -flto -lm

Stargate Digital Audio Workstation

Stargate is an open-source, cross-platform digital audio workstation (DAW) software package with "a unique and carefully curated experience" with scalability from old systems up through modern multi-core systems. Stargate is GPLv3 licensed and makes use of Qt5 (PyQt5) for its user-interface. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgRender Ratio, More Is BetterStargate Digital Audio Workstation 22.11.5Sample Rate: 192000 - Buffer Size: 1024Znver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-5120.64541.29081.93622.58163.227SE +/- 0.009628, N = 3SE +/- 0.014903, N = 3SE +/- 0.003508, N = 3SE +/- 0.000776, N = 32.7837172.8200792.8672332.8683731. (CXX) g++ options: -lpthread -lsndfile -lm -O3 -march=native -ffast-math -funroll-loops -fstrength-reduce -fstrict-aliasing -finline-functions

simdjson

This is a benchmark of SIMDJSON, a high performance JSON parser. SIMDJSON aims to be the fastest JSON parser and is used by projects like Microsoft FishStore, Yandex ClickHouse, Shopify, and others. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgGB/s, More Is Bettersimdjson 2.0Throughput Test: DistinctUserIDZnver3Znver4Znver3 + AVX-512Znver4 + Prefer AVX-512246810SE +/- 0.02, N = 3SE +/- 0.04, N = 3SE +/- 0.04, N = 3SE +/- 0.07, N = 36.466.536.616.86-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CXX) g++ options: -O3 -flto

OpenBenchmarking.orgGB/s, More Is Bettersimdjson 2.0Throughput Test: PartialTweetsZnver4 + Prefer AVX-512Znver4Znver3 + AVX-512Znver3246810SE +/- 0.05, N = 3SE +/- 0.03, N = 3SE +/- 0.03, N = 3SE +/- 0.03, N = 36.436.636.686.74-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -flto

OpenBenchmarking.orgGB/s, More Is Bettersimdjson 2.0Throughput Test: TopTweetZnver3Znver3 + AVX-512Znver4 + Prefer AVX-512Znver4246810SE +/- 0.05, N = 3SE +/- 0.02, N = 3SE +/- 0.07, N = 3SE +/- 0.05, N = 36.616.736.916.96-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto

OpenSSL

OpenSSL is an open-source toolkit that implements SSL (Secure Sockets Layer) and TLS (Transport Layer Security) protocols. This test profile makes use of the built-in "openssl speed" benchmarking capabilities. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgverify/s, More Is BetterOpenSSL 3.0Algorithm: RSA4096Znver4 + Prefer AVX-512Znver3 + AVX-512Znver3Znver4600K1200K1800K2400K3000KSE +/- 10617.65, N = 3SE +/- 1641.03, N = 3SE +/- 228.34, N = 3SE +/- 140.73, N = 32924586.42935372.12938488.52939503.5-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native1. (CC) gcc options: -pthread -m64 -O3 -flto -lssl -lcrypto -ldl

OpenBenchmarking.orgsign/s, More Is BetterOpenSSL 3.0Algorithm: RSA4096Znver4 + Prefer AVX-512Znver3Znver4Znver3 + AVX-51210K20K30K40K50KSE +/- 151.57, N = 3SE +/- 22.49, N = 3SE +/- 8.03, N = 3SE +/- 0.44, N = 344301.844435.344490.144499.3-march=native-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -pthread -m64 -O3 -flto -lssl -lcrypto -ldl

GraphicsMagick

This is a test of GraphicsMagick with its OpenMP implementation that performs various imaging tests on a sample 6000x4000 pixel JPEG image. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: SharpenZnver3Znver4 + Prefer AVX-512Znver3 + AVX-512Znver430060090012001500SE +/- 10.73, N = 3SE +/- 13.17, N = 3SE +/- 1.15, N = 3SE +/- 6.56, N = 31285131413211359-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: RotateZnver3 + AVX-512Znver3Znver4 + Prefer AVX-512Znver4150300450600750SE +/- 3.84, N = 3SE +/- 1.20, N = 3SE +/- 1.00, N = 3SE +/- 1.73, N = 3605645656673-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native-march=native1. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.38Operation: EnhancedZnver3Znver4 + Prefer AVX-512Znver3 + AVX-512Znver45001000150020002500SE +/- 19.19, N = 3SE +/- 5.00, N = 3SE +/- 14.95, N = 3SE +/- 2.00, N = 31837215022082234-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CC) gcc options: -fopenmp -O3 -flto -ljbig -ltiff -lfreetype -ljpeg -lXext -lSM -lICE -lX11 -llzma -lz -lm -lpthread

simdjson

This is a benchmark of SIMDJSON, a high performance JSON parser. SIMDJSON aims to be the fastest JSON parser and is used by projects like Microsoft FishStore, Yandex ClickHouse, Shopify, and others. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgGB/s, More Is Bettersimdjson 2.0Throughput Test: KostyaZnver3 + AVX-512Znver4Znver4 + Prefer AVX-512Znver30.9451.892.8353.784.725SE +/- 0.01, N = 3SE +/- 0.02, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 34.164.174.194.20-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native-march=znver31. (CXX) g++ options: -O3 -flto

JPEG XL Decoding libjxl

The JPEG XL Image Coding System is designed to provide next-generation JPEG image capabilities with JPEG XL offering better image quality and compression over legacy JPEG. This test profile is suited for JPEG XL decode performance testing to PNG output file, the pts/jpexl test is for encode performance. The JPEG XL encoding/decoding is done using the libjxl codebase. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterJPEG XL Decoding libjxl 0.7CPU Threads: 1Znver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver41122334455SE +/- 0.25, N = 3SE +/- 0.12, N = 3SE +/- 0.12, N = 3SE +/- 0.12, N = 347.0248.2148.2248.64

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: IP Shapes 1D - Data Type: u8s8f32 - Engine: CPUZnver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-51248121620SE +/- 0.11, N = 3SE +/- 0.34, N = 12SE +/- 0.24, N = 15SE +/- 0.42, N = 1214.8214.6214.2914.25MIN: 9.81-march=znver3 - MIN: 8.56MIN: 8.38-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 6.051. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

simdjson

This is a benchmark of SIMDJSON, a high performance JSON parser. SIMDJSON aims to be the fastest JSON parser and is used by projects like Microsoft FishStore, Yandex ClickHouse, Shopify, and others. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgGB/s, More Is Bettersimdjson 2.0Throughput Test: LargeRandomZnver4 + Prefer AVX-512Znver3Znver3 + AVX-512Znver40.28580.57160.85741.14321.429SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 31.241.251.251.27-march=native-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CXX) g++ options: -O3 -flto

Kripke

Kripke is a simple, scalable, 3D Sn deterministic particle transport code. Its primary purpose is to research how data layout, programming paradigms and architectures effect the implementation and performance of Sn transport. Kripke is developed by LLNL. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgThroughput FoM, More Is BetterKripke 1.2.4Znver4 + Prefer AVX-512Znver4Znver3Znver3 + AVX-51260M120M180M240M300MSE +/- 2000945.99, N = 3SE +/- 3293132.01, N = 15SE +/- 2950522.72, N = 15SE +/- 3107958.27, N = 12254648533261562280263812847271735708-march=native-march=native-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -fopenmp

WebP Image Encode

This is a test of Google's libwebp with the cwebp image encode utility and using a sample 6000x4000 pixel JPEG image as the input. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterWebP Image Encode 1.2.4Encode Settings: Quality 100, Lossless, Highest CompressionZnver4Znver4 + Prefer AVX-512Znver3Znver3 + AVX-5120.13050.2610.39150.5220.6525SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 30.570.580.580.58-march=native -ljpeg-march=native -ljpeg -ltiff-march=znver3 -lgif -ltiff-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -ljpeg -ltiff1. (CC) gcc options: -fvisibility=hidden -O3 -flto -lm -lpng16

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: IP Shapes 3D - Data Type: bf16bf16bf16 - Engine: CPUZnver3Znver3 + AVX-512Znver4 + Prefer AVX-512Znver41.0182.0363.0544.0725.09SE +/- 0.04256, N = 15SE +/- 0.03566, N = 9SE +/- 0.04436, N = 15SE +/- 0.04728, N = 154.524404.336584.333664.26600-march=znver3 - MIN: 3.03-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 2.98MIN: 2.77MIN: 2.831. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

Stargate Digital Audio Workstation

Stargate is an open-source, cross-platform digital audio workstation (DAW) software package with "a unique and carefully curated experience" with scalability from old systems up through modern multi-core systems. Stargate is GPLv3 licensed and makes use of Qt5 (PyQt5) for its user-interface. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgRender Ratio, More Is BetterStargate Digital Audio Workstation 22.11.5Sample Rate: 96000 - Buffer Size: 1024Znver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-5120.9931.9862.9793.9724.965SE +/- 0.006484, N = 3SE +/- 0.016378, N = 3SE +/- 0.013873, N = 3SE +/- 0.011844, N = 34.3312304.3564464.4082904.4133791. (CXX) g++ options: -lpthread -lsndfile -lm -O3 -march=native -ffast-math -funroll-loops -fstrength-reduce -fstrict-aliasing -finline-functions

Cpuminer-Opt

Cpuminer-Opt is a fork of cpuminer-multi that carries a wide range of CPU performance optimizations for measuring the potential cryptocurrency mining performance of the CPU/processor with a wide variety of cryptocurrencies. The benchmark reports the hash speed for the CPU mining performance for the selected cryptocurrency. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: DeepcoinZnver4Znver3 + AVX-512Znver4 + Prefer AVX-512Znver340K80K120K160K200KSE +/- 81.72, N = 3SE +/- 880.18, N = 3SE +/- 1703.12, N = 5SE +/- 218.28, N = 3159147160157162242164993-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver31. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

SVT-AV1

This is a benchmark of the SVT-AV1 open-source video encoder/decoder. SVT-AV1 was originally developed by Intel as part of their Open Visual Cloud / Scalable Video Technology (SVT). Development of SVT-AV1 has since moved to the Alliance for Open Media as part of upstream AV1 development. SVT-AV1 is a CPU-based multi-threaded video encoder for the AV1 video format with a sample YUV video file. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgFrames Per Second, More Is BetterSVT-AV1 1.4Encoder Mode: Preset 4 - Input: Bosphorus 4KZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-5121.22022.44043.66064.88086.101SE +/- 0.012, N = 3SE +/- 0.037, N = 3SE +/- 0.015, N = 3SE +/- 0.043, N = 35.3605.3745.3925.423-march=znver3-march=znver3 -mavx512cd -mavx512vl -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -march=native -mno-avx -mavx2 -mavx512f -mavx512bw -mavx512dq

libavif avifenc

This is a test of the AOMedia libavif library testing the encoding of a JPEG image to AV1 Image Format (AVIF). Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is Betterlibavif avifenc 0.11Encoder Speed: 2Znver3 + AVX-512Znver4Znver3Znver4 + Prefer AVX-512816243240SE +/- 0.15, N = 3SE +/- 0.26, N = 3SE +/- 0.25, N = 3SE +/- 0.19, N = 334.2334.1033.9133.90-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CXX) g++ options: -O3 -fPIC -flto -lm

Cpuminer-Opt

Cpuminer-Opt is a fork of cpuminer-multi that carries a wide range of CPU performance optimizations for measuring the potential cryptocurrency mining performance of the CPU/processor with a wide variety of cryptocurrencies. The benchmark reports the hash speed for the CPU mining performance for the selected cryptocurrency. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: Quad SHA-256, PyriteZnver3Znver4Znver3 + AVX-512Znver4 + Prefer AVX-512500K1000K1500K2000K2500KSE +/- 5899.09, N = 3SE +/- 10306.20, N = 3SE +/- 22716.96, N = 3SE +/- 25368.27, N = 41378987225106722647472323995-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

PJSIP

PJSIP is a free and open source multimedia communication library written in C language implementing standard based protocols such as SIP, SDP, RTP, STUN, TURN, and ICE. It combines signaling protocol (SIP) with rich multimedia framework and NAT traversal functionality into high level API that is portable and suitable for almost any type of systems ranging from desktops, embedded systems, to mobile handsets. This test profile is making use of pjsip-perf with both the client/server on teh system. More details on the PJSIP benchmark at https://www.pjsip.org/high-performance-sip.htm Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgResponses Per Second, More Is BetterPJSIP 2.11Method: OPTIONS, StatelessZnver4Znver3 + AVX-512Znver4 + Prefer AVX-512Znver370K140K210K280K350KSE +/- 2531.21, N = 3SE +/- 3613.75, N = 15SE +/- 5240.59, N = 15SE +/- 4433.83, N = 12335767336615336791336885-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver31. (CC) gcc options: -lavformat -lavcodec -lswscale -lavutil -lstdc++ -lopus -lssl -lcrypto -lm -lrt -lpthread -lasound -O3 -flto

Cpuminer-Opt

Cpuminer-Opt is a fork of cpuminer-multi that carries a wide range of CPU performance optimizations for measuring the potential cryptocurrency mining performance of the CPU/processor with a wide variety of cryptocurrencies. The benchmark reports the hash speed for the CPU mining performance for the selected cryptocurrency. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: GarlicoinZnver3Znver4 + Prefer AVX-512Znver4Znver3 + AVX-51216K32K48K64K80KSE +/- 66.92, N = 3SE +/- 295.35, N = 3SE +/- 461.75, N = 3SE +/- 742.93, N = 349523721307241372837-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: LBC, LBRY CreditsZnver3Znver4Znver3 + AVX-512Znver4 + Prefer AVX-512200K400K600K800K1000KSE +/- 3568.28, N = 3SE +/- 829.54, N = 3SE +/- 1800.04, N = 3SE +/- 12123.53, N = 3497020106548710677431085827-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: scryptZnver3Znver3 + AVX-512Znver4 + Prefer AVX-512Znver410002000300040005000SE +/- 0.45, N = 3SE +/- 1.99, N = 3SE +/- 1.22, N = 3SE +/- 0.45, N = 32959.154763.914782.744790.11-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: Triple SHA-256, OnecoinZnver3Znver4 + Prefer AVX-512Znver4Znver3 + AVX-512700K1400K2100K2800K3500KSE +/- 4313.22, N = 3SE +/- 22336.69, N = 3SE +/- 26496.25, N = 3SE +/- 26057.45, N = 33255253330121733066433323680-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: x25xZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-5122K4K6K8K10KSE +/- 17.26, N = 3SE +/- 15.17, N = 3SE +/- 74.42, N = 3SE +/- 90.39, N = 36116.977941.388042.888217.70-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: SkeincoinZnver3Znver3 + AVX-512Znver4 + Prefer AVX-512Znver4400K800K1200K1600K2000KSE +/- 5094.00, N = 3SE +/- 24878.62, N = 3SE +/- 7108.55, N = 3SE +/- 10323.18, N = 31414047200499020093672014770-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

OpenBenchmarking.orgkH/s, More Is BetterCpuminer-Opt 3.20.3Algorithm: MagiZnver3 + AVX-512Znver4Znver4 + Prefer AVX-512Znver32K4K6K8K10KSE +/- 50.94, N = 3SE +/- 47.28, N = 3SE +/- 64.85, N = 3SE +/- 51.27, N = 38355.758440.798467.248490.73-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native-march=znver31. (CXX) g++ options: -O3 -flto -lcurl -lz -lpthread -lssl -lcrypto -lgmp

QuantLib

QuantLib is an open-source library/framework around quantitative finance for modeling, trading and risk management scenarios. QuantLib is written in C++ with Boost and its built-in benchmark used reports the QuantLib Benchmark Index benchmark score. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMFLOPS, More Is BetterQuantLib 1.21Znver4Znver4 + Prefer AVX-512Znver3 + AVX-512Znver37001400210028003500SE +/- 1.99, N = 3SE +/- 1.71, N = 3SE +/- 2.74, N = 3SE +/- 4.52, N = 33096.93112.63114.93120.51. (CXX) g++ options: -O3 -march=native -rdynamic

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Deconvolution Batch shapes_1d - Data Type: u8s8f32 - Engine: CPUZnver4 + Prefer AVX-512Znver3Znver4Znver3 + AVX-5120.21470.42940.64410.85881.0735SE +/- 0.009537, N = 3SE +/- 0.009578, N = 5SE +/- 0.012915, N = 3SE +/- 0.011238, N = 40.9541400.9470130.9371640.936374MIN: 0.78-march=znver3 - MIN: 0.79MIN: 0.77-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.791. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

GPAW

GPAW is a density-functional theory (DFT) Python code based on the projector-augmented wave (PAW) method and the atomic simulation environment (ASE). Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is BetterGPAW 22.1Input: Carbon NanotubeZnver3 + AVX-512Znver4 + Prefer AVX-512Znver4Znver3510152025SE +/- 0.07, N = 3SE +/- 0.28, N = 3SE +/- 0.15, N = 3SE +/- 0.26, N = 422.8222.8222.3521.72-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native-march=znver31. (CC) gcc options: -shared -fwrapv -O2 -O3 -flto -lxc -lblas -lmpi

SVT-AV1

This is a benchmark of the SVT-AV1 open-source video encoder/decoder. SVT-AV1 was originally developed by Intel as part of their Open Visual Cloud / Scalable Video Technology (SVT). Development of SVT-AV1 has since moved to the Alliance for Open Media as part of upstream AV1 development. SVT-AV1 is a CPU-based multi-threaded video encoder for the AV1 video format with a sample YUV video file. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgFrames Per Second, More Is BetterSVT-AV1 1.4Encoder Mode: Preset 12 - Input: Bosphorus 4KZnver4 + Prefer AVX-512Znver3 + AVX-512Znver4Znver350100150200250SE +/- 1.78, N = 3SE +/- 3.02, N = 15SE +/- 4.58, N = 15SE +/- 5.39, N = 15196.80206.68210.39222.90-march=znver3 -mavx512cd -mavx512vl -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -march=native -flto -mno-avx -mavx2 -mavx512f -mavx512bw -mavx512dq

JPEG XL Decoding libjxl

The JPEG XL Image Coding System is designed to provide next-generation JPEG image capabilities with JPEG XL offering better image quality and compression over legacy JPEG. This test profile is suited for JPEG XL decode performance testing to PNG output file, the pts/jpexl test is for encode performance. The JPEG XL encoding/decoding is done using the libjxl codebase. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterJPEG XL Decoding libjxl 0.7CPU Threads: AllZnver3 + AVX-512Znver3Znver4 + Prefer AVX-512Znver460120180240300SE +/- 1.24, N = 3SE +/- 1.29, N = 3SE +/- 0.30, N = 3SE +/- 1.41, N = 3266.53269.59272.31277.20

Kvazaar

This is a test of Kvazaar as a CPU-based H.265/HEVC video encoder written in the C programming language and optimized in Assembly. Kvazaar is the winner of the 2016 ACM Open-Source Software Competition and developed at the Ultra Video Group, Tampere University, Finland. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgFrames Per Second, More Is BetterKvazaar 2.1Video Input: Bosphorus 4K - Video Preset: Ultra FastZnver4 + Prefer AVX-512Znver3 + AVX-512Znver3Znver420406080100SE +/- 0.73, N = 15SE +/- 0.64, N = 15SE +/- 0.62, N = 3SE +/- 1.01, N = 374.9876.7678.0178.69-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native1. (CC) gcc options: -pthread -ftree-vectorize -fvisibility=hidden -O3 -flto -lpthread -lm -lrt

GROMACS

The GROMACS (GROningen MAchine for Chemical Simulations) molecular dynamics package testing with the water_GMX50 data. This test profile allows selecting between CPU and GPU-based GROMACS builds. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgNs Per Day, More Is BetterGROMACS 2022.1Implementation: MPI CPU - Input: water_GMX50_bareZnver3Znver3 + AVX-512Znver4 + Prefer AVX-512Znver4510152025SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.02, N = 318.2319.0919.4419.49-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto

Coremark

This is a test of EEMBC CoreMark processor benchmark. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgIterations/Sec, More Is BetterCoremark 1.0CoreMark Size 666 - Iterations Per SecondZnver3Znver4Znver4 + Prefer AVX-512Znver3 + AVX-5122M4M6M8M10MSE +/- 62802.17, N = 3SE +/- 11460.97, N = 3SE +/- 83462.96, N = 3SE +/- 2823.14, N = 37640546.277694653.907861097.867871273.93-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -O2 -O3 -flto -lrt" -lrt

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Deconvolution Batch shapes_1d - Data Type: bf16bf16bf16 - Engine: CPUZnver3 + AVX-512Znver4Znver3Znver4 + Prefer AVX-5120.53141.06281.59422.12562.657SE +/- 0.00584, N = 3SE +/- 0.00839, N = 3SE +/- 0.00532, N = 3SE +/- 0.00909, N = 32.361662.333632.318582.29880-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 1.92MIN: 1.95-march=znver3 - MIN: 1.91MIN: 1.921. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

SVT-AV1

This is a benchmark of the SVT-AV1 open-source video encoder/decoder. SVT-AV1 was originally developed by Intel as part of their Open Visual Cloud / Scalable Video Technology (SVT). Development of SVT-AV1 has since moved to the Alliance for Open Media as part of upstream AV1 development. SVT-AV1 is a CPU-based multi-threaded video encoder for the AV1 video format with a sample YUV video file. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgFrames Per Second, More Is BetterSVT-AV1 1.4Encoder Mode: Preset 13 - Input: Bosphorus 4KZnver4Znver4 + Prefer AVX-512Znver3 + AVX-512Znver350100150200250SE +/- 3.33, N = 15SE +/- 3.20, N = 15SE +/- 2.81, N = 15SE +/- 4.56, N = 12196.12208.64210.43219.57-march=znver3 -mavx512cd -mavx512vl -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -march=native -flto -mno-avx -mavx2 -mavx512f -mavx512bw -mavx512dq

Liquid-DSP

LiquidSDR's Liquid-DSP is a software-defined radio (SDR) digital signal processing library. This test profile runs a multi-threaded benchmark of this SDR/DSP library focused on embedded platform usage. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgsamples/s, More Is BetterLiquid-DSP 2021.01.31Threads: 384 - Buffer Length: 256 - Filter Length: 57Znver3Znver4Znver4 + Prefer AVX-512Znver3 + AVX-5122000M4000M6000M8000M10000MSE +/- 10066445.91, N = 3SE +/- 7356025.50, N = 3SE +/- 7218802.61, N = 3SE +/- 2403700.85, N = 311176000000112493333331127066666711301666667-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -O3 -flto -pthread -lm -lc -lliquid

OpenBenchmarking.orgsamples/s, More Is BetterLiquid-DSP 2021.01.31Threads: 256 - Buffer Length: 256 - Filter Length: 57Znver3Znver4Znver4 + Prefer AVX-512Znver3 + AVX-5122000M4000M6000M8000M10000MSE +/- 24004606.04, N = 3SE +/- 5651843.36, N = 3SE +/- 8533658.85, N = 3SE +/- 4697162.26, N = 39735666667978950000098098000009813700000-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -O3 -flto -pthread -lm -lc -lliquid

OpenBenchmarking.orgsamples/s, More Is BetterLiquid-DSP 2021.01.31Threads: 128 - Buffer Length: 256 - Filter Length: 57Znver3Znver4 + Prefer AVX-512Znver4Znver3 + AVX-5121500M3000M4500M6000M7500MSE +/- 6548367.06, N = 3SE +/- 6145549.43, N = 3SE +/- 3192874.01, N = 3SE +/- 4658445.14, N = 36940833333697763333369908666676999233333-march=znver3-march=native-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CC) gcc options: -O3 -flto -pthread -lm -lc -lliquid

miniBUDE

MiniBUDE is a mini application for the the core computation of the Bristol University Docking Engine (BUDE). This test profile currently makes use of the OpenMP implementation of miniBUDE for CPU benchmarking. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgBillion Interactions/s, More Is BetterminiBUDE 20210901Implementation: OpenMP - Input Deck: BM2Znver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-51260120180240300SE +/- 1.75, N = 3SE +/- 0.73, N = 3SE +/- 0.73, N = 3SE +/- 1.06, N = 3264.15264.28265.52266.161. (CC) gcc options: -std=c99 -Ofast -ffast-math -fopenmp -march=native -lm

OpenBenchmarking.orgGFInst/s, More Is BetterminiBUDE 20210901Implementation: OpenMP - Input Deck: BM2Znver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-51214002800420056007000SE +/- 43.78, N = 3SE +/- 18.35, N = 3SE +/- 18.27, N = 3SE +/- 26.43, N = 36603.686607.096638.016653.861. (CC) gcc options: -std=c99 -Ofast -ffast-math -fopenmp -march=native -lm

WebP Image Encode

This is a test of Google's libwebp with the cwebp image encode utility and using a sample 6000x4000 pixel JPEG image as the input. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterWebP Image Encode 1.2.4Encode Settings: Quality 100, LosslessZnver4Znver3 + AVX-512Znver4 + Prefer AVX-512Znver30.33080.66160.99241.32321.654SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 31.451.451.471.47-march=native -ljpeg-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -ljpeg -ltiff-march=native -ljpeg -ltiff-march=znver3 -lgif -ltiff1. (CC) gcc options: -fvisibility=hidden -O3 -flto -lm -lpng16

SMHasher

SMHasher is a hash function tester supporting various algorithms and able to make use of AVX and other modern CPU instruction set extensions. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgcycles/hash, Fewer Is BetterSMHasher 2022-08-22Hash: MeowHash x86_64 AES-NIZnver4Znver3 + AVX-512Znver3Znver4 + Prefer AVX-5121020304050SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 344.9844.9744.9644.95-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects

OpenBenchmarking.orgMiB/sec, More Is BetterSMHasher 2022-08-22Hash: MeowHash x86_64 AES-NIZnver4Znver3Znver3 + AVX-512Znver4 + Prefer AVX-51212K24K36K48K60KSE +/- 7.42, N = 3SE +/- 9.98, N = 3SE +/- 10.52, N = 3SE +/- 9.47, N = 354272.1454281.9454284.6554297.04-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects
OpenBenchmarking.orgcycles/hash, Fewer Is BetterSMHasher 2022-08-22Hash: MeowHash x86_64 AES-NIZnver4Znver3 + AVX-512Znver3Znver4 + Prefer AVX-5121020304050SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 344.9844.9744.9644.95-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Convolution Batch Shapes Auto - Data Type: bf16bf16bf16 - Engine: CPUZnver4 + Prefer AVX-512Znver4Znver3Znver3 + AVX-5120.10020.20040.30060.40080.501SE +/- 0.000431, N = 3SE +/- 0.001884, N = 3SE +/- 0.002906, N = 3SE +/- 0.004441, N = 150.4455320.4438910.4427480.431299MIN: 0.34MIN: 0.37-march=znver3 - MIN: 0.34-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.331. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Matrix Multiply Batch Shapes Transformer - Data Type: bf16bf16bf16 - Engine: CPUZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-5120.1150.230.3450.460.575SE +/- 0.006281, N = 3SE +/- 0.006958, N = 3SE +/- 0.004211, N = 3SE +/- 0.003087, N = 30.5109390.5048420.5005790.483979-march=znver3 - MIN: 0.39-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.39MIN: 0.39MIN: 0.391. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

SVT-AV1

This is a benchmark of the SVT-AV1 open-source video encoder/decoder. SVT-AV1 was originally developed by Intel as part of their Open Visual Cloud / Scalable Video Technology (SVT). Development of SVT-AV1 has since moved to the Alliance for Open Media as part of upstream AV1 development. SVT-AV1 is a CPU-based multi-threaded video encoder for the AV1 video format with a sample YUV video file. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgFrames Per Second, More Is BetterSVT-AV1 1.4Encoder Mode: Preset 8 - Input: Bosphorus 4KZnver3 + AVX-512Znver4Znver4 + Prefer AVX-512Znver320406080100SE +/- 0.52, N = 3SE +/- 0.33, N = 3SE +/- 0.32, N = 3SE +/- 0.46, N = 392.9693.7694.3495.12-march=znver3 -mavx512cd -mavx512vl -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -mavx512f -mavx512bw -mavx512dq -flto -march=native -mno-avx -mavx2

Kvazaar

This is a test of Kvazaar as a CPU-based H.265/HEVC video encoder written in the C programming language and optimized in Assembly. Kvazaar is the winner of the 2016 ACM Open-Source Software Competition and developed at the Ultra Video Group, Tampere University, Finland. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgFrames Per Second, More Is BetterKvazaar 2.1Video Input: Bosphorus 4K - Video Preset: MediumZnver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver41428425670SE +/- 0.10, N = 3SE +/- 0.46, N = 3SE +/- 0.36, N = 3SE +/- 0.52, N = 363.8064.0564.2764.35-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CC) gcc options: -pthread -ftree-vectorize -fvisibility=hidden -O3 -flto -lpthread -lm -lrt

OpenBenchmarking.orgFrames Per Second, More Is BetterKvazaar 2.1Video Input: Bosphorus 4K - Video Preset: Very FastZnver4Znver3 + AVX-512Znver4 + Prefer AVX-512Znver31632486480SE +/- 0.83, N = 3SE +/- 0.87, N = 4SE +/- 0.92, N = 3SE +/- 0.28, N = 370.7170.7170.8172.01-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver31. (CC) gcc options: -pthread -ftree-vectorize -fvisibility=hidden -O3 -flto -lpthread -lm -lrt

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: IP Shapes 3D - Data Type: u8s8f32 - Engine: CPUZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-5120.20430.40860.61290.81721.0215SE +/- 0.011617, N = 3SE +/- 0.002136, N = 3SE +/- 0.004516, N = 3SE +/- 0.004696, N = 30.9079410.9023030.8866420.881303-march=znver3 - MIN: 0.75-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.75MIN: 0.76MIN: 0.741. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

SMHasher

SMHasher is a hash function tester supporting various algorithms and able to make use of AVX and other modern CPU instruction set extensions. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgcycles/hash, Fewer Is BetterSMHasher 2022-08-22Hash: FarmHash32 x86_64 AVXZnver4 + Prefer AVX-512Znver3 + AVX-512Znver3Znver4612182430SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 326.5026.4926.4926.49-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects

OpenBenchmarking.orgMiB/sec, More Is BetterSMHasher 2022-08-22Hash: FarmHash32 x86_64 AVXZnver3 + AVX-512Znver4 + Prefer AVX-512Znver4Znver39K18K27K36K45KSE +/- 1.84, N = 3SE +/- 1.44, N = 3SE +/- 0.95, N = 3SE +/- 0.13, N = 340559.3340563.9640565.0740565.72-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -flto -march=native -flto=auto -fno-fat-lto-objects
OpenBenchmarking.orgcycles/hash, Fewer Is BetterSMHasher 2022-08-22Hash: FarmHash32 x86_64 AVXZnver4 + Prefer AVX-512Znver3 + AVX-512Znver3Znver4612182430SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 326.5026.4926.4926.49-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Deconvolution Batch shapes_3d - Data Type: u8s8f32 - Engine: CPUZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-5120.06290.12580.18870.25160.3145SE +/- 0.002168, N = 15SE +/- 0.003530, N = 3SE +/- 0.001957, N = 12SE +/- 0.001864, N = 30.2796550.2769650.2747690.274662-march=znver3 - MIN: 0.23-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.24MIN: 0.24MIN: 0.241. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

ASTC Encoder

ASTC Encoder (astcenc) is for the Adaptive Scalable Texture Compression (ASTC) format commonly used with OpenGL, OpenGL ES, and Vulkan graphics APIs. This test profile does a coding test of both compression/decompression. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMT/s, More Is BetterASTC Encoder 4.0Preset: MediumZnver4 + Prefer AVX-512Znver3Znver4Znver3 + AVX-512110220330440550SE +/- 0.56, N = 3SE +/- 6.17, N = 3SE +/- 3.36, N = 13SE +/- 5.68, N = 3420.70459.69493.22511.52-march=native-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -pthread

OpenBenchmarking.orgMT/s, More Is BetterASTC Encoder 4.0Preset: ExhaustiveZnver3Znver4Znver3 + AVX-512Znver4 + Prefer AVX-5123691215SE +/- 0.03, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.05, N = 312.2212.9413.0313.09-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CXX) g++ options: -O3 -flto -pthread

WebP Image Encode

This is a test of Google's libwebp with the cwebp image encode utility and using a sample 6000x4000 pixel JPEG image as the input. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterWebP Image Encode 1.2.4Encode Settings: Quality 100, Highest CompressionZnver3 + AVX-512Znver4 + Prefer AVX-512Znver4Znver30.8191.6382.4573.2764.095SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 33.113.233.253.64-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -ljpeg -ltiff-march=native -ljpeg -ltiff-march=native -ljpeg-march=znver3 -lgif -ltiff1. (CC) gcc options: -fvisibility=hidden -O3 -flto -lm -lpng16

SMHasher

SMHasher is a hash function tester supporting various algorithms and able to make use of AVX and other modern CPU instruction set extensions. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgcycles/hash, Fewer Is BetterSMHasher 2022-08-22Hash: t1ha0_aes_avx2 x86_64Znver4 + Prefer AVX-512Znver3Znver3 + AVX-512Znver4510152025SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 320.8120.5320.5320.52-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects

OpenBenchmarking.orgMiB/sec, More Is BetterSMHasher 2022-08-22Hash: t1ha0_aes_avx2 x86_64Znver3Znver4Znver4 + Prefer AVX-512Znver3 + AVX-51220K40K60K80K100KSE +/- 8.85, N = 3SE +/- 20.00, N = 3SE +/- 11.55, N = 3SE +/- 19.58, N = 3102351.87102354.57102399.74102403.92-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -flto -march=native -flto=auto -fno-fat-lto-objects
OpenBenchmarking.orgcycles/hash, Fewer Is BetterSMHasher 2022-08-22Hash: t1ha0_aes_avx2 x86_64Znver4 + Prefer AVX-512Znver3Znver3 + AVX-512Znver4510152025SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 320.8120.5320.5320.52-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi1. (CXX) g++ options: -O3 -march=native -flto -flto=auto -fno-fat-lto-objects

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Convolution Batch Shapes Auto - Data Type: u8s8f32 - Engine: CPUZnver3 + AVX-512Znver3Znver4 + Prefer AVX-512Znver40.08840.17680.26520.35360.442SE +/- 0.004423, N = 3SE +/- 0.000630, N = 3SE +/- 0.003460, N = 3SE +/- 0.000892, N = 30.3928750.3927650.3884830.380486-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.28-march=znver3 - MIN: 0.28MIN: 0.28MIN: 0.281. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

ASTC Encoder

ASTC Encoder (astcenc) is for the Adaptive Scalable Texture Compression (ASTC) format commonly used with OpenGL, OpenGL ES, and Vulkan graphics APIs. This test profile does a coding test of both compression/decompression. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMT/s, More Is BetterASTC Encoder 4.0Preset: ThoroughZnver3Znver3 + AVX-512Znver4Znver4 + Prefer AVX-512306090120150SE +/- 0.10, N = 3SE +/- 0.06, N = 3SE +/- 0.12, N = 3SE +/- 0.08, N = 3116.09117.69118.74118.99-march=znver3-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=native1. (CXX) g++ options: -O3 -flto -pthread

libavif avifenc

This is a test of the AOMedia libavif library testing the encoding of a JPEG image to AV1 Image Format (AVIF). Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is Betterlibavif avifenc 0.11Encoder Speed: 6, LosslessZnver3 + AVX-512Znver4 + Prefer AVX-512Znver3Znver41.01572.03143.04714.06285.0785SE +/- 0.010, N = 3SE +/- 0.064, N = 3SE +/- 0.027, N = 3SE +/- 0.052, N = 44.5144.4624.4374.398-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CXX) g++ options: -O3 -fPIC -flto -lm

miniBUDE

MiniBUDE is a mini application for the the core computation of the Bristol University Docking Engine (BUDE). This test profile currently makes use of the OpenMP implementation of miniBUDE for CPU benchmarking. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgBillion Interactions/s, More Is BetterminiBUDE 20210901Implementation: OpenMP - Input Deck: BM1Znver4 + Prefer AVX-512Znver3Znver4Znver3 + AVX-51250100150200250SE +/- 2.51, N = 4SE +/- 2.52, N = 3SE +/- 2.07, N = 3SE +/- 2.13, N = 3211.02214.08214.50214.631. (CC) gcc options: -std=c99 -Ofast -ffast-math -fopenmp -march=native -lm

OpenBenchmarking.orgGFInst/s, More Is BetterminiBUDE 20210901Implementation: OpenMP - Input Deck: BM1Znver4 + Prefer AVX-512Znver3Znver4Znver3 + AVX-51212002400360048006000SE +/- 62.65, N = 4SE +/- 62.89, N = 3SE +/- 51.63, N = 3SE +/- 53.26, N = 35275.435351.945362.545365.611. (CC) gcc options: -std=c99 -Ofast -ffast-math -fopenmp -march=native -lm

libavif avifenc

This is a test of the AOMedia libavif library testing the encoding of a JPEG image to AV1 Image Format (AVIF). Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is Betterlibavif avifenc 0.11Encoder Speed: 10, LosslessZnver3 + AVX-512Znver3Znver4 + Prefer AVX-512Znver40.82061.64122.46183.28244.103SE +/- 0.038, N = 3SE +/- 0.016, N = 3SE +/- 0.017, N = 3SE +/- 0.012, N = 33.6473.5813.5723.541-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver3-march=native-march=native1. (CXX) g++ options: -O3 -fPIC -flto -lm

ACES DGEMM

This is a multi-threaded DGEMM benchmark. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgGFLOP/s, More Is BetterACES DGEMM 1.0Sustained Floating-Point RateZnver4Znver3 + AVX-512Znver4 + Prefer AVX-512Znver31632486480SE +/- 0.10, N = 3SE +/- 0.28, N = 3SE +/- 0.07, N = 3SE +/- 0.06, N = 370.0570.1970.3070.38-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=znver31. (CC) gcc options: -O3 -march=native -fopenmp -flto

oneDNN

This is a test of the Intel oneDNN as an Intel-optimized library for Deep Neural Networks and making use of its built-in benchdnn functionality. The result is the total perf time reported. Intel oneDNN was formerly known as DNNL (Deep Neural Network Library) and MKL-DNN before being rebranded as part of the Intel oneAPI toolkit. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgms, Fewer Is BetteroneDNN 3.0Harness: Deconvolution Batch shapes_3d - Data Type: bf16bf16bf16 - Engine: CPUZnver4Znver3Znver4 + Prefer AVX-512Znver3 + AVX-5120.14810.29620.44430.59240.7405SE +/- 0.005902, N = 3SE +/- 0.006925, N = 4SE +/- 0.001081, N = 3SE +/- 0.003927, N = 30.6581010.6563420.6500670.647087MIN: 0.53-march=znver3 - MIN: 0.54MIN: 0.53-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi - MIN: 0.561. (CXX) g++ options: -O3 -march=native -flto -fopenmp -msse4.1 -fPIC -pie -ldl -lpthread

LAMMPS Molecular Dynamics Simulator

LAMMPS is a classical molecular dynamics code, and an acronym for Large-scale Atomic/Molecular Massively Parallel Simulator. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgns/day, More Is BetterLAMMPS Molecular Dynamics Simulator 23Jun2022Model: Rhodopsin ProteinZnver3Znver4Znver3 + AVX-512Znver4 + Prefer AVX-5121224364860SE +/- 0.43, N = 3SE +/- 0.21, N = 3SE +/- 0.40, N = 10SE +/- 0.48, N = 351.4651.5951.8252.29-march=znver3-march=native-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native1. (CXX) g++ options: -O3 -flto -lm -ldl

libavif avifenc

This is a test of the AOMedia libavif library testing the encoding of a JPEG image to AV1 Image Format (AVIF). Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgSeconds, Fewer Is Betterlibavif avifenc 0.11Encoder Speed: 6Znver3 + AVX-512Znver4Znver3Znver4 + Prefer AVX-5120.54141.08281.62422.16562.707SE +/- 0.005, N = 3SE +/- 0.014, N = 3SE +/- 0.010, N = 3SE +/- 0.002, N = 32.4062.3472.3312.317-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi-march=native-march=znver3-march=native1. (CXX) g++ options: -O3 -fPIC -flto -lm

WebP Image Encode

This is a test of Google's libwebp with the cwebp image encode utility and using a sample 6000x4000 pixel JPEG image as the input. Learn more via the OpenBenchmarking.org test page.

OpenBenchmarking.orgMP/s, More Is BetterWebP Image Encode 1.2.4Encode Settings: Quality 100Znver3 + AVX-512Znver3Znver4Znver4 + Prefer AVX-5123691215SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.03, N = 3SE +/- 0.01, N = 311.3811.4811.5411.54-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -ljpeg -ltiff-march=znver3 -lgif -ltiff-march=native -ljpeg-march=native -ljpeg -ltiff1. (CC) gcc options: -fvisibility=hidden -O3 -flto -lm -lpng16

OpenBenchmarking.orgMP/s, More Is BetterWebP Image Encode 1.2.4Encode Settings: DefaultZnver3 + AVX-512Znver4Znver4 + Prefer AVX-512Znver3510152025SE +/- 0.07, N = 3SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.02, N = 318.8518.9518.9718.99-march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -ljpeg -ltiff-march=native -ljpeg-march=native -ljpeg -ltiff-march=znver3 -lgif -ltiff1. (CC) gcc options: -fvisibility=hidden -O3 -flto -lm -lpng16

99 Results Shown

JPEG XL libjxl
CP2K Molecular Dynamics
oneDNN
GraphicsMagick
JPEG XL libjxl
SecureMark
oneDNN
GraphicsMagick
oneDNN
Zstd Compression:
  19, Long Mode - Decompression Speed
  19, Long Mode - Compression Speed
GraphicsMagick
OpenSSL
oneDNN
LAMMPS Molecular Dynamics Simulator
JPEG XL libjxl:
  JPEG - 90
  PNG - 90
Ngspice:
  C2670
  C7552
Zstd Compression:
  19 - Decompression Speed
  19 - Compression Speed
GraphicsMagick
PJSIP
oneDNN
PJSIP
libavif avifenc
Stargate Digital Audio Workstation
simdjson:
  DistinctUserID
  PartialTweets
  TopTweet
OpenSSL:
  RSA4096:
    verify/s
    sign/s
GraphicsMagick:
  Sharpen
  Rotate
  Enhanced
simdjson
JPEG XL Decoding libjxl
oneDNN
simdjson
Kripke
WebP Image Encode
oneDNN
Stargate Digital Audio Workstation
Cpuminer-Opt
SVT-AV1
libavif avifenc
Cpuminer-Opt
PJSIP
Cpuminer-Opt:
  Garlicoin
  LBC, LBRY Credits
  scrypt
  Triple SHA-256, Onecoin
  x25x
  Skeincoin
  Magi
QuantLib
oneDNN
GPAW
SVT-AV1
JPEG XL Decoding libjxl
Kvazaar
GROMACS
Coremark
oneDNN
SVT-AV1
Liquid-DSP:
  384 - 256 - 57
  256 - 256 - 57
  128 - 256 - 57
miniBUDE:
  OpenMP - BM2:
    Billion Interactions/s
    GFInst/s
WebP Image Encode
SMHasher:
  MeowHash x86_64 AES-NI:
    cycles/hash
    MiB/sec
oneDNN:
  Convolution Batch Shapes Auto - bf16bf16bf16 - CPU
  Matrix Multiply Batch Shapes Transformer - bf16bf16bf16 - CPU
SVT-AV1
Kvazaar:
  Bosphorus 4K - Medium
  Bosphorus 4K - Very Fast
oneDNN
SMHasher:
  FarmHash32 x86_64 AVX:
    cycles/hash
    MiB/sec
oneDNN
ASTC Encoder:
  Medium
  Exhaustive
WebP Image Encode
SMHasher:
  t1ha0_aes_avx2 x86_64:
    cycles/hash
    MiB/sec
oneDNN
ASTC Encoder
libavif avifenc
miniBUDE:
  OpenMP - BM1:
    Billion Interactions/s
    GFInst/s
libavif avifenc
ACES DGEMM
oneDNN
LAMMPS Molecular Dynamics Simulator
libavif avifenc
WebP Image Encode:
  Quality 100
  Default

Znver4

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=native -flto" CFLAGS="-O3 -march=native -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Testing initiated at 3 January 2023 08:40 by user phoronix.

Znver4 + Prefer AVX-512

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=native -mprefer-vector-width=512 -flto" CFLAGS="-O3 -march=native -mprefer-vector-width=512 -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Testing initiated at 3 January 2023 18:42 by user phoronix.

Znver3

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=znver3 -flto" CFLAGS="-O3 -march=znver3 -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Testing initiated at 4 January 2023 05:51 by user phoronix.

Znver3 + AVX-512

Processor: 2 x AMD EPYC 9654 96-Core @ 3.71GHz (192 Cores / 384 Threads), Motherboard: AMD Titanite_4G (RTI1002E BIOS), Chipset: AMD Device 14a4, Memory: 1520GB, Disk: 800GB INTEL SSDPF21Q800GB, Graphics: ASPEED, Monitor: VGA HDMI, Network: Broadcom NetXtreme BCM5720 PCIe

OS: Ubuntu 23.04, Kernel: 5.19.0-21-generic (x86_64), Desktop: GNOME Shell 43.1, Display Server: X Server 1.21.1.4, Vulkan: 1.3.224, Compiler: GCC 13.0.0 20230103, File-System: ext4, Screen Resolution: 1920x1080

Kernel Notes: Transparent Huge Pages: madvise
Environment Notes: CXXFLAGS="-O3 -march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -flto" CFLAGS="-O3 -march=znver3 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -flto"
Compiler Notes: --disable-multilib --enable-checking=release
Processor Notes: Scaling Governor: amd-pstate performance (Boost: Enabled) - CPU Microcode: 0xa10110d
Python Notes: Python 3.10.9
Security Notes: itlb_multihit: Not affected + l1tf: Not affected + mds: Not affected + meltdown: Not affected + mmio_stale_data: Not affected + retbleed: 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 Retpolines IBPB: conditional IBRS_FW STIBP: always-on RSB filling PBRSB-eIBRS: Not affected + srbds: Not affected + tsx_async_abort: Not affected

Testing initiated at 4 January 2023 13:46 by user phoronix.