Open-Source NVIDIA Driver Approaches Stable State
It is not part of this testing or readily available yet, also landing soon is going to be a new libdrm for Nouveau that will be more stable and faster. Nouveau developers have reported Nexuiuz as being as much as 100% faster than the current libdrm.
With Nouveau leaving the kernel's staging area I have carried out benchmarks of many NVIDIA GeForce graphics cards comparing the Nouveau driver to the proprietary NVIDIA Linux graphics driver. For the most part Nouveau was in good shape just as it has been for the past few months. With each graphics card re-clocking was attempted to put the card in its highest performance state like is done automatically by the NVIDIA driver when experiencing load, but for a few cards it was still problematic.
Of the tested graphics cards this week, the few that were problematic in one way or another are noted below.
GeForce 8500GT - This graphics card was being unstable with the Linux 3.4 development kernel and Mesa Git master in most OpenGL tests. The GeForce 8500GT was running fine in a few OpenGL tests, but for some test profiles (oddly, it seemed to often be with the not-very-demanding ioquake3 games) that the system would exhibit stability issues. The stability issues happened regardless of clock speed.
GeForce 8600GTS - The 8600GTS graphics card tested only had one performance level, but regardless it would lock-up most times when running Lightsmark and Xonotic.
GeForce 8800GT - The NVIDIA GeForce 8800GT would freeze immediately upon re-clocking to its higher performance levels. This happened every time with the latest Nouveau Linux code so for this benchmarking the 8800GT had to be left at its boot speed of 399MHz core, 810MHz shader, and 399MHz memory rather than 600MHz core, 1350MHz shader, and 700MHz memory.
GeForce 9600GSO - Re-clocking would cause stability problems for this GPU, although a few months ago it was working better.
GeForce 9800GT - For the 9800GT the re-clocking process would be successful, though it would immediately set the graphics card's fan to running at 100% throttle, which for this particular MSI model was obnoxiously noisy. In addition, there would be some on-screen memory corruption within the Ubuntu Unity desktop and an occasional stability problem during OpenGL with workloads like Doom 3 and Lightsmark.
GeForce GT 240 - Immediately upon mode-setting there would be the "tiling corruption" issue that I have long experienced for this graphics card on Nouveau. I do not recall if this graphics card has ever worked well with Nouveau in past years, but this GPU has been extremely problematic at least in recent times for the open-source driver.
GeForce GTX 460 - For the higher-end GeForce GTX 460 Fermi graphics card, re-clocking was successful and didn't cause stability problems... However, while the higher performance levels were shown, when it came to actually re-clocking the graphics card only the core and shader clocks were being changed. The memory re-clocking on the Linux 3.4 kernel for this graphics card never ended up going into affect but would be left at its (slow) memory speed that it was left at by this video BIOS on boot.
GeForce GT 520 - This Fermi graphics card but reportedly with some Kepler mode-setting attributes had constant screen flashing when mode-setting.
GeForce GTX 550 Ti - The GTX 550 Ti had the same problem as the GeForce GTX 460 with the core and shader re-clocking but the video memory failing to change.