For those deciding between the Nouveau (the open-source default NVIDIA driver on Ubuntu) and the proprietary NVIDIA Linux graphics driver that can easily be installed from Ubuntu's package repository, here's some new benchmarks. Overall though, these new benchmarks aren't terribly surprising... NVIDIA's binary driver still largely wins by a landslide, in large part because the Nouveau driver still lacks proper GPU re-clocking support.
Nouveau re-clocking is a nightmare and has long been a work-in-progress. For some older GeForce graphics cards, re-clocking the GPU core, shader, and memory frequencies can be carried out successfully. However, for a vast majority of the supported NVIDIA hardware, it's hit-or-miss and can easily lead to on-screen corruption and stability issues. For newer GeForce 400/500/600 (Fermi and Kepler) generations, the support is currently non-existent. Thus, for a vast majority of the Nouveau users out there, you're running your hardware at a much lower speed than what it's rated for and it's leading to crippling results.
The benchmarks in this article are just from the Lenovo ThinkPad W510 with NVIDIA Quadro FX880M. This GT2xx class GPU is stuck running with a 405MHz core and 324MHz memory frequency over the 550MHz rated clock speed. With higher-end models, the difference between the boot and rated clock speeds is generally much steeper. The "out of the box" Ubuntu 13.04 performance was compared between Nouveau and NVIDIA on this laptop.
The benchmark results in full can be found within 1304112-UT-NVIDIAVSN62.
The results, sadly, aren't surprising. The Nouveau Gallium3D drivers are becoming in fairly good shape compared to the other open-source Mesa/Gallium3D drivers, but the lack of re-clocking within the Nouveau DRM driver seems to be the most pressing problem at the moment.
Sadly, there isn't too much hope on the horizon with the Linux 3.10 kernel.
There are more thorough benchmarks from a wide-range of graphics processors on both drivers coming up next week on Phoronix in a featured, multi-page article.
View more results at OpenBenchmarking.org.