The benchmarking for this more extensive 3.12 DRM testing was completely different from the Ivy Bridge Extreme setup used for delivering the original RV770 results on Saturday. The system used for this article included an Intel Core i7 4770K "Haswell" CPU with OCZ SSD and 16GB of RAM. The software stack was comprised of Ubuntu 13.10 x86_64 with the mainline vanilla 3.11/3.12 kernels, Unity 7.1.1, xf86-video-ati 7.2.0, Mesa 10.0.0-devel git-975c6ce, and LLVM 3.3. For each of the graphics cards tested, the only change made between runs was swapping between the Linux 3.11.0 and Linux 3.12 Git kernels, as obtained from the Ubuntu mainline kernel PPA on Saturday, 12 October.
The selection of Linux OpenGL games used for this benchmarking was based upon those available via the Phoronix Test Suite with OpenBenchmarking.org. As far as why no Steam games were used, see the explanation in this article.
(Ignore the reported clock-speed differences between Linux 3.11 and Linux 3.12. The Intel CPU was running at the same frequencies between 3.11 and 3.12 but now with the new kernel the cpufreq driver is reporting the Turbo frequency rather than the base frequency.)
As said in the original Radeon Linux 3.12 performance article, an updated look of Linux 3.12 + Mesa 10.0 vs. the AMD Catalyst binary driver will arrive in the coming days. Today's article was unplanned to come so soon until after seeing the forum feedback and that even AMD's Alex Deucher isn't even sure why the Linux 3.12 kernel performance is so much higher over Linux 3.11.
The Phoronix Test Suite is capable of automatically bisecting code in spotting performance regressions and we've publicly demonstrated this several times over the years. Perhaps the most widespread help that our benchmarking software's auto-bisecting support has been of help for the Linux kernel was in finding the major Linux kernel power regression back in 2011. For isolating what commit(s) caused this impressive performance boost, the auto-bisecting feature will be called into action once again.
The bisected results will come in the day or so... As Phoronix.com is entirely funded by advertisements and site subscriptions, please disable AdBlock when viewing this site and/or please subscribe to Phoronix Premium or consider a PayPal contribution. All of this testing is done single-handedly by myself while writing literally thousands of articles per year for Phoronix. Just the testing for today's article ended up being twenty hours of work between Saturday and Sunday to complete this massive comparison and was my unexpected way to spend the weekend, on top of my usual seven-day-per-week workload. Your support and donations are appreciated to continue this work especially considering how tough it is for Linux sites. Beyond that, AMD hasn't even been sending out hardware samples lately for Linux testing so the recent AMD GPUs I've actually had to purchase retail. Thank you, I hope to have the bisected information published tomorrow.
This major performance change in the Linux kernel that came as completely unexpected to the upstream developers and can't be easily explained further advances another one of my automated-testing initiatives that I've been pushing for and spearheading for years: daily and per-commit testing of the Linux kernel with real-world workloads. I've done daily Linux kernel benchmarks for over a year to show off Phoromatic, an enterprise component to the Phoronix Test Suite for remotely managing PTS clients and setting up routine test processes. That daily kernel tracker was successful in finding performance regressions during its time in operation before temporarily being halted due to associated costs in operating the systems, etc. As I announced a few days prior to finding the Radeon 3.12 performance changes, I do have plans as well to restart daily and per-commit tests of the Linux kernel, Mesa, compilers, etc. Had this been in place already with an AMD system, the DRM performance changes would have been spotted in real-time. Your feedback and support is appreciated.
Without further ado, here's the results for the Linux 3.12 Radeon DRM performance between the ten different graphics cards: Radeon HD 3850, HD 4670, HD 4830, HD 4890, HD 5830, HD 6450, HD 6570, HD 6450, HD 6570, HD 6770, HD 6870, and HD 6950.