Starting out the last week of July's Linux benchmarking on Phoronix is a fresh comparison of several NVIDIA GeForce graphics cards when comparing the performance of the latest open-source Nouveau driver against the latest NVIDIA proprietary Linux graphics driver. While the Kepler cards now support GPU re-clocking, the results aren't quite ideal yet.
For testing the open-source Nouveau driver stack we used the Linux 3.16 development kernel on Ubuntu 14.04 LTS x86_64 with the Oibaf PPA to provide Mesa 10.3-devel and other updated user-space components. With Linux 3.16 there is basic Nouveau re-clocking support for Kepler. As covered in that earlier article for re-clocking GeForce 600/700 series hardware, for most Kepler GPUs the Nouveau DRM can't currently re-clock to the highest performance state (0f) but is generally limited to a mid-range state like 0a. The only graphics card tested that could fully re-clock was the GeForce GTX 650 while the other GPUs would cause problems when attempting the highest performance state.
The proprietary driver used for testing was the fresh NVIDIA 340.24 release while also pulling back the kernel version (but keeping the same configuration) for managing compatibility with the NVIDIA (and AMD Catalyst, with the other upcoming results) proprietary drivers. The same system configuration and other components were used throughout testing. The graphics cards tested included:
- MSI GeForce 9800GT 512MB
- NVIDIA GeForce GTX 460 768MB
- eVGA GeForce GT 520 1024MB
- Zotac GeForce GT 610 1024MB
- MSI GeForce GTX 650 1024MB
- NVIDIA GeForce GTX 680 2048MB
- NVIDIA GeForce GTX 760 2048MB
All of the OpenGL benchmarking was handled by the Phoronix Test Suite. For those interested in a comparison of the performance-per-Watt and power consumption as tested this weekend for the open-source drivers, the separate open vs. closed thermal/power efficiency article will be published in the coming days. The Radeon Gallium3D vs. AMD Catalyst Linux comparison is also being worked on for publishing this week.