Regressed R9 290 Isn't Affected By Issue When Using AMDGPU DRM Driver
Published earlier today was AMDGPU vs. Radeon DRM benchmarks on DRM-Next for GCN 1.0/1.1 graphics cards that have experimental support available via Kconfig switches to enable this alternative open-source DRM driver over the long-standing Radeon driver. Here are some fresh Git benchmarks using the R9 290 "Hawaii" GPU that's been in a regressed state on the default Radeon driver.
Since Linux 4.7, my Radeon R9 290 as well as the Hawaii-derived cards of various other Linux users (including the equivalent in the Rx 300 series) have been affected by slow performance but it doesn't seem to affect all such GPUs. The issue was narrowed down for some users, including a Mesa fix. Due to the troubled R9 290 support recently, I ran these tests after writing the earlier article using my various other GCN 1.0 and 1.1 GPUs.
Well, with DRM-Next of the new feature code going into Linux 4.9, the regression remains for my R9 290 when also paired with Mesa 12.1-dev Git via the Padoka PPA... But on the bright side, the issue doesn't happen when using the AMDGPU DRM driver!
Quite different from the other AMDGPU vs. Radeon benchmarks published this morning, but only because the Radeon DRM driver is regressed for this card. Chances are if the card wasn't borked with that driver, the performance would be close to the same.
Since Linux 4.7, my Radeon R9 290 as well as the Hawaii-derived cards of various other Linux users (including the equivalent in the Rx 300 series) have been affected by slow performance but it doesn't seem to affect all such GPUs. The issue was narrowed down for some users, including a Mesa fix. Due to the troubled R9 290 support recently, I ran these tests after writing the earlier article using my various other GCN 1.0 and 1.1 GPUs.
Well, with DRM-Next of the new feature code going into Linux 4.9, the regression remains for my R9 290 when also paired with Mesa 12.1-dev Git via the Padoka PPA... But on the bright side, the issue doesn't happen when using the AMDGPU DRM driver!
Quite different from the other AMDGPU vs. Radeon benchmarks published this morning, but only because the Radeon DRM driver is regressed for this card. Chances are if the card wasn't borked with that driver, the performance would be close to the same.
17 Comments