AMD Radeon R9 290: Linux 4.0 vs. Linux 4.1 Git

[phoronix]

Phoronix: AMD Radeon R9 290: Linux 4.0 vs. Linux 4.1 Git

Now that the Linux 4.1 kernel is starting to stabilize, I just started some Radeon DRM open-source graphics tests from this new kernel. So far I did some preliminary tests with the AMD Radeon R9 290 graphics card...

http://www.phoronix.com/scan.php?pag...9-290-DRM-Test

[Pontostroy]

Why 290x so bad in comparison with the 280x? llvm 3.6 or something else?

http://openbenchmarking.org/result/1...DE-R280XMESA85

[smitty3268]

Why 290x so bad in comparison with the 280x? llvm 3.6 or something else?

http://openbenchmarking.org/result/1...DE-R280XMESA85

Yeah, it's probably llvm 3.7 vs 3.6. There are a lot of added optimizations in 3.7.

[tomtomme]

Yeah, it's probably llvm 3.7 vs 3.6. There are a lot of added optimizations in 3.7.

maybe
but michael tested 2560x1600 (says the spec box - but not the graphs - wierd) and you tested fullHD

[smitty3268]

maybe
but michael tested 2560x1600 (says the spec box - but not the graphs - wierd) and you tested fullHD

Oh yeah, missed that. Why do the charts all say 1920x1080? openbenchmarking bug? Or is he running 1920x1080 tests on a desktop with a 2560x1600 monitor? Windowed vs full screen tests?

[tomtomme]

Oh yeah, missed that. Why do the charts all say 1920x1080? openbenchmarking bug? Or is he running 1920x1080 tests on a desktop with a 2560x1600 monitor?

exactly my thought / question...Michael?

[Marc Driftmeyer]

For the record, a lot of added support in LLVM-3.7 for the R9 290 series has been added, that is missing in LLVM-3.6. Michael should be testing against LLVM-3.7 now that LLVM-3.6.1 is near release.

[Michael]

exactly my thought / question...Michael?

It's exactly as the PTS data shows... 2560x1600 desktop but with 1920x1080 fullscreen apps, since when testing the native 2560x1600 of the panel, there's usually complaints that more people use 1080p....

[xeekei]

It's exactly as the PTS data shows... 2560x1600 desktop but with 1920x1080 fullscreen apps, since when testing the native 2560x1600 of the panel, there's usually complaints that more people use 1080p....

I know it would effectively double your work, but testing on both would be good. AMD cards (at least on Windows) tend to take a smaller hit from more pixels than Nvidia cards.

[eydee]

It's exactly as the PTS data shows... 2560x1600 desktop but with 1920x1080 fullscreen apps, since when testing the native 2560x1600 of the panel, there's usually complaints that more people use 1080p....

You're doing it right. Omitting the fact that the monitor is capable of higher resolutions could decrease confusion though.