Proprietary vs. Linux Git, Mesa 11.2-devel, DRI3 For R600g/RadeonSI
Following this week's OpenGL 4.1 R600g benchmarking with that newly-enabled OpenGL 4 support, I set out to run a larger hardware comparison on both the R600g and RadeonSI drivers as part of our year-end 2015 Linux benchmarking. In this article are tests of seven AMD Radeon graphics cards tested on the proprietary driver compared to the latest open-source driver stack -- with extra steps of enabling DRI3 rendering and also using the latest AMDGPU PowerPlay code.
The graphics cards tested for this latest open vs. closed-source driver comparison were the Radeon HD 6870, HD 6950, HD 7950, R9 285, R9 290, R7 370, and R9 Fury. On the binary blob side was Catalyst 15.9 for having official HD 6000 series support still and not seeing any real performance difference out of the Radeon Software Crimson Edition compared to what officially ships as the Ubuntu 15.10 display driver.
On the open-source driver side I was using Alex Deucher's latest AMDGPU PowerPlay kernel branch as of yesterday for allowing the best support for the new AMD Tonga and Fiji GPUs. That PowerPlay branch is based off the Linux 4.4-rc3 kernel. Also enabled was the Padoka PPA to provide Mesa 11.2-devel and LLVM 3.8 SVN as of Friday. Additionally, DRI3 (Direct Rendering Infrastructure 3) was manually enabled via the X.Org configuration for ensuring best possible performance. Last minute change: While writing this article, I decided to drop the results for the R9 285 and R9 Fury for now as it seems the re-clocking support on this PowerPlay support is currently borked or just acting very strangely. I will follow-up with some AMDGPU Fury Fiji/Tonga tests once that re-clocking issue is sorted through. Additionally, those two graphics cards on AMDGPU were only exposing OpenGL 3.0 atop this Mesa stack.
For this OpenGL comparison I first ran some Steam games and other OpenGL Linux tests at 1080p followed by some 4K results for the higher-end graphics cards. All of this benchmarking was done in a fully-automated and reproducible manner using the open-source Phoronix Test Suite benchmarking software. Now let's go see how the performance of the AMD proprietary driver compares to this very fresh open-source driver code.