Total War: WARHAMMER - AMD RadeonSI Linux Performance

[Michael]

If we can even find someone still seeing the problem who could provide more info that would help - the bug ticket activity has dried up since all of the recent reports of performance regression seem to have been things like having a dpm=0 setting left over from earlier experimentation. Will talk to the devs and find out what next steps should be... if Michael is still seeing the regression maybe we can work something out.

If you can confirm that no one has been yet to reproduce internally, feel free to email me if any questions or other feedback from internal, I can probably bisect it in the next few days... People telling me I should take a break on Thanksgiving or whatever, I guess I can sit back and drink some beers while watching PTS bisect the kernel as my sort of fun break...

While PTS automates it, just hadn't done it since I am usually running the same system for all of my graphics tests since it's the only easy one to access for GPU swapping in the top rack of a 24U rack, unfortunately no sliders on any of my cases, so try to only swap the GPU in the top system. And then pretty much constantly that system is running tests for future articles as opposed to bisecting, since I keep all my systems racked these days.

[duby229]

If you can confirm that no one has been yet to reproduce internally, feel free to email me if any questions or other feedback from internal, I can probably bisect it in the next few days... People telling me I should take a break on Thanksgiving or whatever, I guess I can sit back and drink some beers while watching PTS bisect the kernel as my sort of fun break...

While PTS automates it, just hadn't done it since I am usually running the same system for all of my graphics tests since it's the only easy one to access for GPU swapping in the top rack of a 24U rack, unfortunately no sliders on any of my cases, so try to only swap the GPU in the top system. And then pretty much constantly that system is running tests for future articles as opposed to bisecting, since I keep all my systems racked these days.

No offense, but there were quite a few of us that mentioned that to you at the time. As soon as you posted about the chasis you were informed about the accessibility problems you were inevitably going to have. It's easy to blow off after the fact, but you were aware at the fact.

[bridgman]

My impression was that sliders cost as much or more than cases these days, so it may not have been such an easy and obvious choice.

If you can confirm that no one has been yet to reproduce internally, feel free to email me if any questions or other feedback from internal, I can probably bisect it in the next few days... People telling me I should take a break on Thanksgiving or whatever, I guess I can sit back and drink some beers while watching PTS bisect the kernel as my sort of fun break...

Great, thanks. Will ask around internally to figure out what next steps should be, particularly what to check to make sure the bisect is useful and doesn't just rediscover the page flip issue. I don't remember what the resolution for that was, might have just been disabling vsync to avoid having frame rate jumping to 30 whenever it drifted below 60.

[Michael]

My impression was that sliders cost as much or more than cases these days, so it may not have been such an easy and obvious choice.

Yep, sliders aren't cheap. It's only on very rare occasions such as this that I regret not having sliders or some more easy solution.

[agd5f]

The pageflipping issue was a bug in mesa that was fixed in mesa. No kernel changes to worry about. Ideally someone who can reproduce the issue would open a new bug to avoid further issue conflation with the pageflipping bug. Not to add further to the issue conflation, but I added some possible things to try on the pageflipping bug comment 59 (https://bugs.freedesktop.org/show_bug.cgi?id=97260#c59). So far, no one has tried them and reported back.

[UbayGd]

Hi guys!

With this config:

- RX470 OC 8GB
- Padoka PPA
- Kernel 4.7 (from amd-staging-4.7 branch)
- export R600_DEBUG='sbcl,hyperz,llvm,sisched,forcedma' in /etc/environment

I running the benchmark included in the game, and I only get ~20 FPS (average) with 1920x1080 screen resolution and high settings.


Is this normal?? :O

[Pontostroy]

>R600_DEBUG='sbcl,hyperz,llvm,sisched,forcedma'
you need only R600_DEBUG=sisched

[UbayGd]

>R600_DEBUG='sbcl,hyperz,llvm,sisched,forcedma'
you need only R600_DEBUG=sisched

I will try this now. Thank you

[UbayGd]

Changing R600_DEBUG='sbcl,hyperz,llvm,sisched,forcedma' to R600_DEBUG=sisched looks like do the same. In 2 tests, 21 and 24 FPS average (1920x1080, High Settings).

[oooverclocker]

I think you should better choose Kernel 4.8 or higher for the RX 400 series. 4.7 just had initial support.