OpenCL ROCm 2.0 vs. AMDGPU-PRO Linux Radeon RX Vega 64 ROCm 2.0 OpenCL versus PAL OpenCL driver in AMDGPU-PRO 18.50. Benchmarks by Michael Larabel for a future article on Phoronix.com. ,,"ROCm 2.0","AMDGPU-PRO 18.50 PAL" Processor,,AMD Ryzen Threadripper 2990WX 32-Core @ 3.00GHz (32 Cores / 64 Threads),AMD Ryzen Threadripper 2990WX 32-Core @ 3.00GHz (32 Cores / 64 Threads) Motherboard,,ASUS ROG ZENITH EXTREME (1601 BIOS),ASUS ROG ZENITH EXTREME (1601 BIOS) Chipset,,AMD Family 17h,AMD Family 17h Memory,,32768MB,32768MB Disk,,16GB Voyager 3.0 + Samsung SSD 970 EVO 500GB,16GB Voyager 3.0 + Samsung SSD 970 EVO 500GB Graphics,,AMD Radeon RX Vega 8GB (1630/945MHz),AMD Radeon RX Vega 8GB (1630/945MHz) Audio,,Realtek ALC1220,Realtek ALC1220 Monitor,,ASUS VP28U,ASUS VP28U Network,,Intel I211 + Qualcomm Atheros QCA6174 802.11ac + Wilocity Wil6200 802.11ad,Intel I211 + Qualcomm Atheros QCA6174 802.11ac + Wilocity Wil6200 802.11ad OS,,Ubuntu 18.04,Ubuntu 18.04 Kernel,,4.15.0-43-generic (x86_64),4.15.0-43-generic (x86_64) Desktop,,GNOME Shell 3.28.3,GNOME Shell 3.28.3 Display Server,,X Server 1.19.6,X Server 1.19.6 Display Driver,,amdgpu 18.0.1,amdgpu 18.1.99 OpenGL,,4.5 Mesa 18.0.5 (LLVM 6.0.0),4.6.13542 Compiler,,GCC 7.3.0,GCC 7.3.0 File-System,,ext4,ext4 Screen Resolution,,3840x2160,3840x2160 ,,"ROCm 2.0","AMDGPU-PRO 18.50 PAL" "cl-mem - Benchmark: Read (GB/s)",HIB,160,398 "clpeak - OpenCL Test: Transfer Bandwidth enqueueWriteBuffer (GBPS)",HIB,45.44,22.49 "Darktable - Test: Server Rack - Acceleration: OpenCL (sec)",LIB,0.23,0.13 "cl-mem - Benchmark: Copy (GB/s)",HIB,221,364 "JuliaGPU - OpenCL Device: GPU (Samples/sec)",HIB,166858816,240293469 "Darktable - Test: Server Room - Acceleration: OpenCL (sec)",LIB,1.97,2.57 "SHOC Scalable HeterOgeneous Computing - Target: OpenCL - Benchmark: FFT SP (GFLOPS)",HIB,1075,863 "PlaidML - FP16: No - Mode: Inference - Network: Inception V3 - Device: OpenCL (Examples/sec)",HIB,113.88,136.38 "SHOC Scalable HeterOgeneous Computing - Target: OpenCL - Benchmark: Triad (GB/s)",HIB,6.69,6.07 "SHOC Scalable HeterOgeneous Computing - Target: OpenCL - Benchmark: MD5 Hash (GHash/s)",HIB,16.46,17.63 "Rodinia - Test: OpenCL Heartwall (sec)",LIB,4.03,3.80 "PlaidML - FP16: No - Mode: Inference - Network: IMDB LSTM - Device: OpenCL (Examples/sec)",HIB,252,263 "Darktable - Test: Masskrug - Acceleration: OpenCL (sec)",LIB,5.70,5.47 "clpeak - OpenCL Test: Single-Precision Float (GFLOPS)",HIB,13053,12528 "SHOC Scalable HeterOgeneous Computing - Target: OpenCL - Benchmark: Texture Read Bandwidth (GB/s)",HIB,441,424 "PlaidML - FP16: No - Mode: Inference - Network: ResNet 50 - Device: OpenCL (Examples/sec)",HIB,225,233 "cl-mem - Benchmark: Write (GB/s)",HIB,384,379 "clpeak - OpenCL Test: Double-Precision Double (GFLOPS)",HIB,833,828 "clpeak - OpenCL Test: Integer Compute INT (GIOPS)",HIB,2497,2486 "clpeak - OpenCL Test: Global Memory Bandwidth (GBPS)",HIB,362,361 "LeelaChessZero - Backend: OpenCL (Nodes/s)",HIB,304, "PlaidML - FP16: No - Mode: Inference - Network: Mobilenet - Device: OpenCL (Examples/sec)",HIB,479,479 "SHOC Scalable HeterOgeneous Computing - Target: OpenCL - Benchmark: Bus Speed Readback (GB/s)",HIB,7.16,7.16 "SHOC Scalable HeterOgeneous Computing - Target: OpenCL - Benchmark: Bus Speed Download (GB/s)",HIB,7.14,7.14 "clpeak - OpenCL Test: Transfer Bandwidth enqueueReadBuffer (GBPS)",HIB,17.06,10.92 "clpeak - OpenCL Test: Kernel Latency (us)",LIB,10.56,41.75 "Darktable - Test: Boat - Acceleration: OpenCL (sec)",LIB,4.48,8.50