AMD & NVIDIA: Open vs. Closed-Source Driver Performance

Continuing on from this weekend's open-source Nouveau vs. closed-source NVIDIA Linux driver performance are results now added in with showing AMD's open-source vs. closed-source driver performance with the same tests.

As a reminder from the earlier article, the open-source NVIDIA tests were done with the Nouveau stack found in Linux 4.10 and Mesa 13.1-dev. With the supported Kepler GPUs re-clocking was manually enabled to the 0f pstate along with enabling NvBoost support, new to Linux 4.10. The NVIDIA binary driver tested was the 375.26 driver. The cards tested were the GeForce GTX 680, GTX 760, GTX 780 Ti, GTX 980, and GTX 980 Ti. Testing was limited of the GTX 900 Maxwell GPUs due to there not yet being re-clocking there on Nouveau and no GTX 1000 Pascal cards were tested since there isn't yet any accelerated open-source driver support.

For the new results on the AMD side were open-source numbers from using the same Linux 4.10 and Mesa 13.1-dev stack as used during the Nouveau testing. The closed-source AMD driver used was their AMDGPU-PRO 16.50 release. The cards tested there were R9 285, RX 480, and R9 Fury for both a mix of older and new cards with all of AMD's current Radeon GPUs being supported by the open-source driver stack so not having to worry about any limitations like there is with Nouveau when it comes to hardware support or re-clocking.

These open-source vs. closed-source GPU linux driver benchmarks for both AMD and NVIDIA were carried out in a fully-automated manner using the open-source Phoronix Test Suite benchmarking software. Only OpenGL graphics tests were carried out due to the open-source NVIDIA (Nouveau) driver not having any Vulkan driver currently. Likewise, no OpenCL tests were conducted since the open-source AMD and NVIDIA stacks both lack adequate OpenCL drivers at the moment.