R9 Fury Performance Gains With Linux 4.5-rc2? I'm Not So Lucky

[phoronix]

Phoronix: R9 Fury Performance Gains With Linux 4.5-rc2? I'm Not So Lucky

Since yesterday's release of Linux 4.5-rc2 there was at least one report of AMDGPU performance improvements with Linux 4.5-rc2 for an R9 Fury "Fiji" graphics card...

http://www.phoronix.com/scan.php?page=news_item&px=R9-Fury-Linux-4.5-rc2-Test

[andrei_me]

Michael have you tried another Linux? Maybe Ubuntu is to blame here? What is different from your setup and from the others that have been able to use Fury with great success on AMDGPU ?

[bug77]

Really Michael, you fell for that "quadrupled fps" line? Cause if you did, I have some penis-enlargement pills to sell you

[Guest]

Phoronix: R9 Fury Performance Gains With Linux 4.5-rc2? I'm Not So Lucky

Since yesterday's release of Linux 4.5-rc2 there was at least one report of AMDGPU performance improvements with Linux 4.5-rc2 for an R9 Fury "Fiji" graphics card...

http://www.phoronix.com/scan.php?pag...x-4.5-rc2-Test

Just a question about the article: what does it mean "latest firmware"?

[Michael]

Michael have you tried another Linux? Maybe Ubuntu is to blame here? What is different from your setup and from the others that have been able to use Fury with great success on AMDGPU ?

Haven't tried other distros since the R9 285 is working great on this same Ubuntu-based stack... So it's not something universally bad about Ubuntu but rather product specific. And not a hardware defect as Catalyst runs great on the same Ubuntu and same hardware.

[Michael]

Just a question about the article: what does it mean "latest firmware"?

The latest firmware blobs as needed for Radeon/AMDGPU usage.

[smitty3268]

Michael, did you ever test this patch that AMD asked you to try?

For those having performance problems with Fiji, please try this patch:
http://people.freedesktop.org/~agd5f...-on-fiji.patch
and see if it helps. On production boards, the vbios loads the MC firmware so there's no need for the driver to do it. Perhaps some boards don't like the MC firmware image the driver is loading. It's probably better to stick with the vbios provided MC firmware anyway since the images were validated with the boards. The driver had to load the MC firmware on early engineering sample boards.

[ernstp]

Here's my Fury results for Xonotic. Also tested with R600_DEBUG=sisched:
http://openbenchmarking.org/result/1...KH-FURY8013634

Updated with added Unigine Heaven:

http://openbenchmarking.org/result/1602017-KH-FURY9420734

[Herem]

Here's my Fury results for Xonotic. Also tested with R600_DEBUG=sisched:
http://openbenchmarking.org/result/1...KH-FURY8013634

Nice results, especially as you had P-State powersave set instead of P-State performance.

If Michael's tests from the other day were indicative you might be able to gain another 6% performance (~190fps) by switching to performance.

[Hibbelharry]

I also gave a spin to the new kernel and get satisfying results, see here:

http://openbenchmarking.org/result/1602019-HA-XONOTIC1370

Ubuntu 16.04 in it's current state, just swapped the kernel with Michaels 4.5rc2 from today. Little bite confusing: Kernel was really that rc2 build from Michael posted today, I don't know why pts gets this wrong. Double checked by uname -a and friends.