OpenCL Radeon Open-Source vs. Catalyst

Radeon R600g and RadeonSI Gallium3D Clover OpenCL benchmarks with clpeak using Linux 4.5 + Mesa 11.2-devel. Compared to Catalyst. Benchmark results by Michael Larabel of Phoronix.com for a future article.

HTML result view exported from: https://openbenchmarking.org/result/1602026-GA-OPENCLAMD14&grs&sor.

ProcessorMotherboardChipsetMemoryDiskGraphicsAudioMonitorNetworkOSKernelDesktopDisplay ServerDisplay DriverOpenGLCompilerFile-SystemScreen ResolutionOpen-SourceCatalyst HD 6870 HD 7950 R9 285 R9 290 R7 370 HD 6870 HD 7950 R9 285 R9 290 R7 370Intel Core i7-5960X @ 3.50GHz (16 Cores)Gigabyte X99-UD4-CFIntel Xeon E7 v3/Xeon16384MB2 x 120GB SanDisk SDSSDA12 + 120GB Samsung SSD 850Sapphire AMD Radeon HD 6870 1024MBRealtek ALC1150DELL P2415QIntel ConnectionUbuntu 15.104.5.0-rc2amdgpu-pp+ (x86_64) 20160201UnityX Server 1.17.2radeon 7.6.993.3 Mesa 11.2.0-devel (padoka PPA) Gallium 0.4GCC 5.2.1 20151010ext43840x2160XFX AMD Radeon HD 7950/8950 OEM / R9 280 3072MB4.1 Mesa 11.2.0-devel (padoka PPA) Gallium 0.4XFX AMD Radeon R9 285/380 2048MBamdgpu 1.0.99XFX AMD Radeon R9 290 4096MBradeon 7.6.99MSI AMD Radeon R7 370 / R9 270/370 OEM 4096MBSapphire AMD Radeon HD 6800 1024MB4.2.0-27-generic (x86_64)fglrx 15.20.34.5.13399XFX AMD Radeon HD 7900 3072MBSupported device 6939 2048MBSupported device 67B1 4096MBMSI AMD Radeon R7 370 4096MBOpenBenchmarking.orgKernel Details- Open-Source: HD 6870, Open-Source: HD 7950, Open-Source: R9 285, Open-Source: R9 290, Open-Source: R7 370: amdgpu.powerplay=1Compiler Details- --build=x86_64-linux-gnu --disable-browser-plugin --disable-vtable-verify --disable-werror --enable-checking=release --enable-clocale=gnu --enable-gnu-unique-object --enable-gtk-cairo --enable-java-awt=gtk --enable-java-home --enable-languages=c,ada,c++,java,go,d,fortran,objc,obj-c++ --enable-libmpx --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-multiarch --enable-multilib --enable-nls --enable-objc-gc --enable-plugin --enable-shared --enable-threads=posix --host=x86_64-linux-gnu --target=x86_64-linux-gnu --with-abi=m64 --with-arch-32=i686 --with-arch-directory=amd64 --with-default-libstdcxx-abi=new --with-multilib-list=m32,m64,mx32 --with-tune=generic -v Processor Details- Scaling Governor: intel_pstate performanceGraphics Details- Open-Source: HD 6870: EXA- Open-Source: HD 7950: GLAMOR- Open-Source: R9 285: GLAMOR- Open-Source: R9 290: GLAMOR- Open-Source: R7 370: GLAMOREnvironment Details- Catalyst: HD 6870, Catalyst: HD 7950, Catalyst: R9 285, Catalyst: R9 290, Catalyst: R7 370: LIBGL_DRIVERS_PATH=/usr/lib/fglrx/dri:/usr/lib/x86_64-linux-gnu/dri:/usr/lib/dri:/usr/lib32/fglrx/dri:/usr/lib/i386-linux-gnu/dri

clpeak: Integer Compute INTclpeak: Double-Precision Doubleclpeak: Transfer Bandwidth enqueueWriteBufferclpeak: Single-Precision Floatclpeak: Transfer Bandwidth enqueueReadBufferclpeak: Kernel LatencyOpen-SourceCatalyst HD 6870 HD 7950 R9 285 R9 290 R7 370 HD 6870 HD 7950 R9 285 R9 290 R7 370182.445.26360.923.58145.85491.62547.0315.722345.1210.40105.54650.80204.7514.823102.6010.09143.391578.01596.5512.324608.559.82118.71418.74131.4915.772031.9110.41110.24199.389.49398.298.8024.90516.74536.8413.882292.5410.4734.03647.08202.7113.213177.629.0026.781591.74598.8114.814721.1810.6437.77415.67128.7314.972095.9110.6735.42OpenBenchmarking.org

clpeak

OpenCL Test: Integer Compute INT

CatalystOpen-SourceOpenBenchmarking.orgGIOPS, More Is BetterclpeakOpenCL Test: Integer Compute INTR9 290R9 285HD 7950R7 370HD 687030060090012001500SE +/- 0.05, N = 3SE +/- 1.77, N = 3SE +/- 21.03, N = 6SE +/- 3.65, N = 3SE +/- 0.02, N = 3SE +/- 0.49, N = 3SE +/- 0.05, N = 3SE +/- 7.09, N = 6SE +/- 0.01, N = 3SE +/- 0.16, N = 31591.74647.08516.74415.67199.381578.01650.80491.62418.74182.44

clpeak

OpenCL Test: Double-Precision Double

CatalystOpen-SourceOpenBenchmarking.orgGFLOPS, More Is BetterclpeakOpenCL Test: Double-Precision DoubleR9 290HD 7950R9 285R7 370130260390520650SE +/- 0.01, N = 3SE +/- 11.37, N = 6SE +/- 0.22, N = 3SE +/- 0.36, N = 3SE +/- 0.10, N = 3SE +/- 16.28, N = 6SE +/- 0.03, N = 3SE +/- 0.00, N = 3598.81536.84202.71128.73596.55547.03204.75131.49

clpeak

OpenCL Test: Transfer Bandwidth enqueueWriteBuffer

Open-SourceCatalystOpenBenchmarking.orgGBPS, More Is BetterclpeakOpenCL Test: Transfer Bandwidth enqueueWriteBufferR7 370HD 7950R9 285R9 290HD 687048121620SE +/- 0.00, N = 3SE +/- 0.05, N = 3SE +/- 0.59, N = 6SE +/- 0.06, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.03, N = 3SE +/- 0.04, N = 3SE +/- 0.19, N = 3SE +/- 0.12, N = 315.7715.7214.8212.325.2614.9714.8113.8813.219.49

clpeak

OpenCL Test: Single-Precision Float

CatalystOpen-SourceOpenBenchmarking.orgGFLOPS, More Is BetterclpeakOpenCL Test: Single-Precision FloatR9 290R9 285HD 7950R7 370HD 687010002000300040005000SE +/- 0.95, N = 3SE +/- 0.24, N = 3SE +/- 44.66, N = 6SE +/- 2.25, N = 3SE +/- 0.00, N = 3SE +/- 2.13, N = 3SE +/- 1.52, N = 3SE +/- 47.92, N = 6SE +/- 0.22, N = 3SE +/- 0.08, N = 34721.183177.622292.542095.91398.294608.553102.602345.122031.91360.92

clpeak

OpenCL Test: Transfer Bandwidth enqueueReadBuffer

CatalystOpen-SourceOpenBenchmarking.orgGBPS, More Is BetterclpeakOpenCL Test: Transfer Bandwidth enqueueReadBufferR7 370R9 290HD 7950R9 285HD 68703691215SE +/- 0.04, N = 3SE +/- 0.13, N = 3SE +/- 0.11, N = 3SE +/- 0.07, N = 3SE +/- 0.08, N = 3SE +/- 0.02, N = 3SE +/- 0.03, N = 3SE +/- 0.30, N = 6SE +/- 0.25, N = 6SE +/- 0.10, N = 610.6710.6410.479.008.8010.4110.4010.099.823.58

clpeak

OpenCL Test: Kernel Latency

CatalystOpen-SourceOpenBenchmarking.orgus, Fewer Is BetterclpeakOpenCL Test: Kernel LatencyHD 6870R9 285HD 7950R7 370R9 290306090120150SE +/- 0.28, N = 3SE +/- 0.38, N = 3SE +/- 0.68, N = 3SE +/- 0.52, N = 5SE +/- 2.12, N = 6SE +/- 1.78, N = 6SE +/- 3.85, N = 6SE +/- 2.08, N = 6SE +/- 14.32, N = 6SE +/- 8.80, N = 624.9026.7834.0335.4237.77105.54110.24118.71143.39145.85

Phoronix Test Suite v10.8.5