The Most Comprehensive AMD Radeon Linux Graphics Comparison

While not covered by this review, there is open-source support for the AMD Fusion APUs, including the latest Llano hardware. This is supported still by the Radeon DRM driver and the R600 Gallium3D driver. See the many Fusion Linux articles I have written on the topic of Fusion APUs since their launch.

Also coming up will be the Radeon HD 7000 "Southern Islands" launch. While the open-source support is coming along, do not expect it at launch day. There will be Catalyst support at launch for Linux users, but the Radeon HD 7000 series support is not expected until the Linux 3.3 kernel or potentially even the Linux 3.4 kernel. At first it might also be limited to kernel mode-setting and the 2D/3D acceleration coming to Gallium3D at a later date. Based upon the current tick of things, I will be happy if there is open-source support for the Southern Islands by the start of the 2012 calendar year, but that may still be a bit optimistic. Fortunately, in another year, with the Radeon HD 8000 series we may see support closer to launch.

Besides needing to wait around for open-source hardware support and a driver that does not match the performance or feature-richness of the Catalyst offering, another current shortcoming of the open-source Radeon driver is the lack of a formal QA (Quality Assurance) process and CI (Continuous Integration) to ward off regressions and show-stopping problems. It has become a very frequent problem for the driver to regress between kernel releases or for the Mesa driver to break for different games as OpenGL changes are made, etc. Intel does have some testing infrastructure for their open-source driver via their China team (some of this testing does involve the use of the Phoronix Test Suite). Meanwhile, the Radeon and Nouveau drivers pretty much rely upon user-generated bug reports.

As talked about at XDC2011 Chicago, I have been working on a bit of a solution, at least for catching performance regressions. I'm assembling a performance monitoring farm for Mesa that will test each driver on a per-commit basis, but even that won't be perfect as I don't have an unlimited power budget for constantly adding new systems nor an unlimited supply of hardware for each generation. Regardless, it should be an improvement over the status quo. There is also kernel DRM testing that will need to be done. All of this work will be powered by OpenBenchmarking.org and the Phoronix Test Suite, which is also used by internally by Advanced Micro Devices for OpenGL QA on their Linux driver and within other parts of AMD and most other companies.