Earlier this week I posted some benchmarks showing the open-source NVIDIA (Nouveau) driver performance on Linux 4.10 with the new NvBoost capability for finally being able to hit the "boost" clock frequencies with Kepler graphics cards when using this reverse-engineered driver. While the manual re-clocking and enabling NvBoost is able to increase the Nouveau driver's performance, how do these results compare to using the closed-source NVIDIA Linux driver? These benchmarks answer that question.
For this article are the best results obtained from that former article when using Nouveau on Linux 4.10 (for the Kepler cards, manually re-clocked to the 0f pstate and then enabling NvBoost=2 but for the newer Maxwell cards there isn't yet re-clocking or NvBoost support) then compared to the proprietary NVIDIA Linux driver, the 375.26 latest stable release for this binary driver.
The cards tested under these latest open-source and closed-source NVIDIA Linux drivers were the GeForce GTX 680, GTX 760, GTX 780 Ti, GTX 980, and GTX 980 Ti. Yep, mostly Kepler coverage given that's where the best Nouveau driver support is currently and then with Maxwell the performance is low due to no open-source re-clocking while for the GeForce GTX 1000 series the 3D acceleration support is currently M.I.A.
On the following pages are these open-source Nouveau (Linux 4.10 + Mesa 13.1-dev) versus NVIDIA 375.26 GPU benchmark results. All tests were done in a fully-automated and reproducible manner using the open-source Phoronix Test Suite benchmarking software.