Nouveau vs. NVIDIA Linux Comparison Shows Shortcomings

Written by Michael Larabel in Display Drivers on 23 April 2013 at 01:58 AM EDT. Page 2 of 8. 61 Comments.

Aside from the re-clocking mess for the Nouveau driver, other problems ran into with this open-source driver on the stack tested were:

- Nouveau regressed on the GeForce 8500GT due to DRM driver errors.

- The NVIDIA GeForce GT 240 continues to fail when it comes to loading the xf86-video-nouveau DDX driver with corruption issues. It's been a long-standing issue where a VT will work fine in a Nouveau KMS-enabled world along with Plymouth, but immediately when the X.Org driver is loaded for this particular GPU things go awry.

- The NVIDIA GeForce GTX 680 graphics card has been tested earlier with Nouveau on Phoronix, but now with this GPU on the Linux 3.9 kernel it was producing endless PGRAPH errors.

- Of the 11 tested NVIDIA GPUs, the only graphics cards where re-clocking was successful were the GeForce 9500GT, 9800GTX, and GT 220.

Sadly, the Nouveau kernel driver seems to regress quite frequently, still making it like a game of Russian Roulette in between major Linux kernel releases. The Nouveau Gallium3D drivers tend to be a bit more reliable and regress less for the OpenGL support.

Ubuntu 13.04 x86_64 was the base operating system used throughout this OpenGL NVIDIA Linux benchmarking. For the open-source driver stack it was made of Mesa 9.2.0-devel git-cb12bf7, Linux 3.9 kernel Git, and xf86-video-nouveau 1.0.7. This was the same software makeup of last week's open-source Radeon driver testing. Likewise, Mesa was built with the "--enable-texture-float" argument for the Gallium3D driver. While Unity/Compiz has improved a lot and doesn't impact the OpenGL gaming performance as much these days for Ubuntu 13.04, Xfce 4.10 was the desktop in use.

The binary NVIDIA Linux graphics driver used was the latest NVIDIA 319.12 Beta Linux graphics driver.

Nouveau vs. NVIDIA Linux OpenGL - April 2013

As with last week's Radeon vs. Catalyst driver comparison, the range of OpenGL benchmarks used for testing were those that work reliably on both the open-source and closed-source drivers and for each can at least deliver meaningful frame-rates. For those wanting to see Source Engine Linux benchmarks, those will come in a separate article when working around automated test issues with the Steam client. All benchmarking was handled in a fully automated and reproducible manner through the Phoronix Test Suite automated testing framework.


Related Articles