A 14-Way Comparison Of NVIDIA vs. Nouveau Drivers

That about covers all of the general details for the various generations of NVIDIA graphics cards being tested. Simply put, the NVIDIA binary Linux driver support is generally spot-on in terms of features and performance to the Windows driver. There are a few exceptions, like the lack of Fermi overclocking support and NVIDIA Optimus support, but overall there is not much more than one could ask from their binary blob.

When it comes to the Nouveau driver support, there is usually a lot to be desired. The performance of the Nouveau driver with Gallium3D is usually much slower than the proprietary driver (though the Radeon Gallium3D situation isn't too different, except when Nouveau fails at re-clocking to the nominal speeds), the support usually comes several months after the hardware launches due to no support from NVIDIA and the clean-room reverse-engineering that must take place on limited manpower, and there's far from being a feature parity to NVIDIA's official Linux driver.

The Nouveau driver can be usable for those that just care about a basic composited Linux desktop for office and Internet tasks, but for gaming and other demanding workloads it is not too viable. Nouveau use-cases are also limited for many due to the lack of reputable power management and fan management support. For mobile users with the Nouveau driver can find themselves with a very warm laptop. Mobile and desktop users can also be annoyed by the Nouveau driver, due to the associated noise level by pushing the fan at its maximum speed by default.

Making matters worse for Nouveau is that due to the lack of official NVIDIA support and the only documentation coming about via community reverse-engineering, is that regressions are all too common in Mesa/Gallium3D and the kernel driver, which leads to the Russian Roulette situation. With frequently testing different hardware and driver combinations, I am commonly hit with show-stopping regressions in the mode-setting becoming either borked or stability issues under OpenGL workloads, which seem to be the most common problems.

This testing was done using the latest Mesa 7.12-devel (Git master), xf86-video-nouveau DDX, and Linux 3.2 kernel as of 7 November. Ubuntu 11.10 x86_64 was the operating system used with the Unity desktop. The latest NVIDIA binary Linux driver was also utilized. The setup used for this NVIDIA graphics card testing was the AMD FX-8150 "Bulldozer" kit. Both drivers were tested in their stock configurations. This includes driver options and the default Nouveau performance levels, etc.

Besides running a variety of OpenGL games/benchmarks that are capable of running on the open-source Mesa/Gallium3D in the stock configuration, additional tests were carried out by the Phoronix Test Suite to look at the CPU usage, GPU core temperature, and system power consumption for the range of NVIDIA graphics cards under both drivers.

Now let's begin with the results...