Forcing The R9 Fury's Performance State For Better AMDGPU Performance?

[phoronix]

Phoronix: Forcing The R9 Fury's Performance State For Better AMDGPU Performance?

With yesterday's Linux 4.5 AMDGPU/Radeon vs. Catalyst OpenGL Performance testing, the loser was the R9 Fury "Fiji" graphics card on the AMDGPU open-source driver with the performance still being miserable. Even with PowerPlay enabled on Linux 4.5, the performance was still poor. But if forcing to a high performance state via sysfs, is the performance any better?..

http://www.phoronix.com/scan.php?page=news_item&px=R9-Fury-High-Perf-State

[jf33]

Michael, did you ask AMD if you could send them your card? Since they can't reproduce the issue yet, they should be very interested. And they should provide you with a working Fury in exchange.

[Hackerpilot]

This may help the AMD guys narrow down the cause of this bug: I have that same model of R9 Fury and the performance is terrible. It's not a re-clocking problem, checking /sys/kernel/debug/dri/0/amdgpu_pm_info while (attempting to) play a game showed the clock speeds being what I expect them to be.

[Michael]

Michael, did you ask AMD if you could send them your card? Since they can't reproduce the issue yet, they should be very interested. And they should provide you with a working Fury in exchange.

No one has ever contacted me about such an exchange; I don't know where a few people got that idea from and are mentioning it.

[Daktyl198]

No one has ever contacted me about such an exchange; I don't know where a few people got that idea from and are mentioning it.

The guy clearly asked if YOU got in contact with AMD, not the other way around...

[bug77]

This is the particular Sapphire R9 Fury I bought last summer after launch. So I'd recommend staying clear of it as an open-source AMD Linux user

I think it goes without saying we'll all stay clear of the particular card you bought
Yeah, I get it, there may be something funny in the BIOS of this model.

[Michael]

The guy clearly asked if YOU got in contact with AMD, not the other way around...

In some other threads or Twitter, there seemed to be the indication some thought the other way around and here he seems to think for AMD that they can easily do such an exchange.

[Michael]

I think it goes without saying we'll all stay clear of the particular card you bought
Yeah, I get it, there may be something funny in the BIOS of this model.

Yep and other person that just commented on this thread in having the same issue has this same model as well.

[BenPope]

We don't have any overlaps on these benchmarks, but mine has been performing much better: http://openbenchmarking.org/result/1...HA-BPPADOKAN35

On the weekend I'll try your kernel and I'll also update my build and give that a go.

Last time I tried passing the kernel parameter I got bad results, instead I just changed the default in the source before building. Hopefully we can shed some light.

If you have suggestions for a run that only has freely available benchmarks, I'll run it too. My native screen res is odd at 3440x1440, but I'm happy to change it if it helps, and of course, is possible.

[Michael]

If you have suggestions for a run that only has freely available benchmarks, I'll run it too. My native screen res is odd at 3440x1440, but I'm happy to change it if it helps, and of course, is possible.

You can run any of my result files. When it tries to isntall any of the Steam-using tests, it will just fail and continue. When it asks you if you want to try reinstalling the tests, just say no and it will skip over them when benchmarking.