Following yesterday's 11-way Talos Principle comparison on NVIDIA using the latest drivers, I proceeded to run some AMD tests... Catalyst had to be left out though since I was just getting a black screen when running it on the box with the proprietary driver enabled. So that left the open-source driver stack: it worked, but wasn't perfect.
I was testing the RadeonSI Gallium3D driver stack from Mesa 11.2-devel + LLVM 3.9 SVN via the Padoka PPA on this Ubuntu 15.10 system. Both the Radeon and AMDGPU kernel drivers were used depending upon the hardware; a Linux 4.5-rc3 kernel with AMDGPU PowerPlay enabled was used for this benchmarking.
The performance is rather poor as shown by the results, but at least it worked. From the time spent with the driver stack and Talos Principle was within the benchmark mode. The rendering for the most part was on-par, but the shadows in particular weren't rendering as well compared to the proprietary NVIDIA driver stack.
A Phoronix reader had commented about his RadeonSI Talos Principle experience while playing the game, "I did play the whole game on an AMD HD7850 with radeonsi driver, it was playable but with a few bugs: shadows were jerky and textures were sometimes missing/very slow to load. The game is a masterpiece and it's still worth playing it."
So here's the 4K RadeonSI performance for this game compared to the NVIDIA results from yesterday:
The performance on the open-source driver is well below the NVIDIA graphics cards and their high-performance binary blob. Even with using Linux 4.5-rc3 with AMDGPU PowerPlay enabled, the R9 285 and R9 Fury were performing not as well as the other GCN GPUs using the mature Radeon DRM driver. But in turn all of the red GPUs weren't close to competing with the NVIDIA Linux results.
There wasn't much of a difference in CPU usage.
To dig through the results more, you can see this Talos Radeon result file with all of the benchmark results. Or there's this combined results merge for the NVIDIA vs. AMD data points on this Ubuntu 15.10 system, etc.