13-Way Radeon AMDGPU-PRO 17.50 vs. NVIDIA Linux OpenCL Compute Comparison

Given this week's release of the big AMDGPU-PRO 17.50 Linux driver update, here are some fresh OpenCL GPU benchmarks comparing the performance of AMD's latest Radeon graphics cards on this newest Linux driver to that of the latest NVIDIA GeForce GPUs on their respective newest driver.

For this testing the AMDGPU-PRO 17.50 driver was used for all the Radeon testing: initially there were plans to do a legacy vs. ROCm comparison now that the AMDGPU-PRO driver makes it easy with the --opencl=legacy and --opencl=rocm options, but at least with 17.50, the ROCm OpenCL driver doesn't end up working on pre-Vega GPUs. But the upstream ROCm source should work for AMD Fiji/Polaris GPUs and newer. So coming up in a follow-up article will be fresh tests using the latest upstream ROCm code aside from these AMDGPU-PRO 17.50 OpenCL tests done today.

On the NVIDIA side was the 387.34 Linux x86_64 driver. The range of graphics cards tested included the:

- Radeon R9 290
- Radeon RX 560
- Radeon RX 580
- Radeon R9 Fury
- Radeon RX Vega 56
- Radeon RX Vega 64
- GeForce GTX 1050
- GeForce GTX 1050 Ti
- GeForce GTX 1060
- GeForce GTX 1070
- GeForce GTX 1070 Ti
- GeForce GTX 1080
- GeForce GTX 1080 Ti

This round of OpenCL testing was done with Ubuntu 16.04.3 LTS on its default Linux 4.10 kernel. In addition to the raw OpenCL performance results, there are also performance-per-dollar results that follow. All of these OpenCL GPGPU benchmarks were carried out in a fully-automated and reproducible manner using the open-source Phoronix Test Suite benchmarking software.