Nouveau Driver Support Still Like Russian Roulette

With the recent Intel Ivy Bridge graphics benchmarks and the newer Radeon Gallium3D results, the plan was also for some updated Nouveau Gallium3D benchmarks too ahead of the release of Linux 3.8 and Mesa 9.1. Unfortunately, the reverse-engineered open-source NVIDIA support is still very much hit or miss.

From mid-January were the Five-Way NVIDIA GeForce Comparison On Nouveau results. Unfortunately though there's been some regressions in the past month. Aside from Linux 3.8 regressions, the Nouveau re-clocking support is still less than desirable. However, I was told earlier this month at FOSDEM there should at least be some progress with Kepler re-clocking to report on soon.

Right now I'm working through the latest Nouveau problems, but for those that have been requesting new results of Nouveau, embedded below are some of the results I have been able to manage thus far. They're results from a NVIDIA GeForce GT 220 that was able to re-clock to 625MHz for its core and 405MHz on the video memory. There were no lock-ups in re-clocking, but on-screen corruption was quite common when clocked to its appropriate frequencies rather than the 399/399MHz boot frequencies. Nouveau from Mesa 9.1 Git as of this week was used as well as the Linux 3.8 development kernel as of this week. Results in full for this GeForce GT 220 benchmarking between Mesa 9.1 and 9.0 Git can be found on OpenBenchmarking.org in 1302135-FO-MESA91GEF03.

The results are mixed with regard to whether Mesa 9.1 is faster than its predecessor. Results from other NVIDIA GeForce graphics cards will come soon as some other issues/regressions with the Nouveau stack are worked out so it's back to playing nicely with more of the test hardware.

Continue with the rest of the data in 1302135-FO-MESA91GEF03.