Early Linux 4.15 AMDGPU Linux Gaming Tests Indicate Some Regressions
Here are some early AMD Radeon Linux gaming benchmarks using the in-development Linux 4.15 kernel. Unfortunately, there are a few performance regressions.
I carried out some quick OpenGL/Vulkan Linux gaming tests with two Radeon GPUs comparing the stable Linux 4.14.0 kernel to the Linux 4.15 Git state as of yesterday, which has in all of the AMDGPU DRM feature updates. With this comparing to 4.14 stable, RX Vega isn't tested in this article since only AMDGPU DC just landed -- those tests are being worked on right now. For 4.14 vs. 4.15 Git, an RX 580 and R9 Fury were used for testing.
This Ubuntu 17.10 system meanwhile had Mesa 17.4-dev via the Padoka PPA. Besides the addition of DC display code, for a run-down on all what's changed for Linux 4.15 when it comes to AMDGPU, see our DRM pull request coverage. There's also an overview of the other 4.15 changes so far.
In most tests, no performance changes to note with Linux 4.15... But not really a surprise considering AMD has been busy beating AMDGPU DC into shape, etc.
In some tests, the frame-rates may be a few frames higher but difficult to call anything definitive.
But in Deus Ex: Mankind Divided, there is an evident regression with both graphics cards on Linux 4.15. Though with it not being widespread, it's not likely any re-clocking issue and so the AMDGPU memory management changes in 4.15 come to mind as something possibly explaining this, but the AMD developers in our forums will likely have their own ideas to share.
The R9 Fury with Unigine Superposition is also slower with Linux 4.15.
That's about it for the AMDGPU changes seen so far with Linux 4.15 while more benchmarks are in the works. Additional results and test details can be found on OpenBenchmarking.org.
I carried out some quick OpenGL/Vulkan Linux gaming tests with two Radeon GPUs comparing the stable Linux 4.14.0 kernel to the Linux 4.15 Git state as of yesterday, which has in all of the AMDGPU DRM feature updates. With this comparing to 4.14 stable, RX Vega isn't tested in this article since only AMDGPU DC just landed -- those tests are being worked on right now. For 4.14 vs. 4.15 Git, an RX 580 and R9 Fury were used for testing.
This Ubuntu 17.10 system meanwhile had Mesa 17.4-dev via the Padoka PPA. Besides the addition of DC display code, for a run-down on all what's changed for Linux 4.15 when it comes to AMDGPU, see our DRM pull request coverage. There's also an overview of the other 4.15 changes so far.
In most tests, no performance changes to note with Linux 4.15... But not really a surprise considering AMD has been busy beating AMDGPU DC into shape, etc.
In some tests, the frame-rates may be a few frames higher but difficult to call anything definitive.
But in Deus Ex: Mankind Divided, there is an evident regression with both graphics cards on Linux 4.15. Though with it not being widespread, it's not likely any re-clocking issue and so the AMDGPU memory management changes in 4.15 come to mind as something possibly explaining this, but the AMD developers in our forums will likely have their own ideas to share.
The R9 Fury with Unigine Superposition is also slower with Linux 4.15.
That's about it for the AMDGPU changes seen so far with Linux 4.15 while more benchmarks are in the works. Additional results and test details can be found on OpenBenchmarking.org.
24 Comments