With this week having delivered our 2014 Catalyst Linux Graphics Benchmarks Year-In-Review that looked at the evolution of the AMD Catalyst Linux driver performance over this year, now it's time to see how the open-source Radeon driver performance has evolved.
As the open-source Linux graphics drivers aren't self-contained like the proprietary drivers but rather have components spread across user-space and kernel-space, for this testing we ended up comparing the performance of Ubuntu 13.10 from the end of 2013 to Ubuntu 14.10 plus a third run of using the latest Git components. With a few AMD GPUs we tested:
- Ubuntu 13.10 out-of-the-box with Mesa 9.2.1 and Linux 3.11 kernel.
- Ubuntu 14.10 with Mesa 10.3.0 and Linux 3.16 kernel.
- The Ubuntu 14.10 installation upgraded to using the Linux 3.18.0 vanilla kernel and employing the paulo-miguel-dias PPA for having the Oibaf packages plus LLVM 3.6 SVN. This provided the latest AMD GPU LLVM back-end of 3.6 SVN along with Mesa 10.5-devel Git and the xf86-video-ati Git.
The graphics cards tested for this article consisted of an AMD Radeon HD 6870, HD 7850, and R9 270X graphics card as an assortment of new and old AMD graphics processors. The Radeon R9 290 series wasn't tested for this article since the open-source Hawaii support wasn't working at the end of 2013 and only got into good shape with the recent Ubuntu 14.10 release. On a similar note, the Radeon R9 285 Tonga GPU wasn't tested for this article since there is no open-source driver support at all until the new AMDGPU kernel DRM driver and other new unified code premieres.
Beyond the open-source numbers for this article and the recent Catalyst comparison, next week will be these three GPUs with the open-source numbers compared to the Catalyst 14.12 Omega driver on Ubuntu 14.10 for an end-of-year comparison between the very latest open and closed-source AMD Linux graphics drivers. There's also more end-of-year results coming for the Intel and NVIDIA Linux graphics code.
Note: All of the same software/hardware settings were used throughout testing. For those out of the loop and concerned by the reported CPU frequency differences on the automated system table, that just comes down to a base vs. turbo frequency reporting difference between the ACPI CPUfreq and P-State scaling drivers used on Ubuntu 13.10 and 14.10 with their respective kernels.
All of the OpenGL benchmarks for this article were run in a fully-automated and reproducible manner using the open-source Phoronix Test Suite benchmarking software. Beyond the performance changes, there were also many new open-source AMD Linux driver features added this year, as outlined at the end of this article.