A Radeon RX 590 Workaround For Linux But With Abysmal Performance

[phoronix]

Phoronix: A Radeon RX 590 Workaround For Linux But With Abysmal Performance

While AMD is able to reproduce the Radeon RX 590 Linux failure and is currently investigating the necessary Linux driver fix(es) for getting this latest Polaris refresh graphics card working correctly, if you already upgraded and don't have the luxury of switching to another graphics card until a solution is in place, there is a workaround to getting the RX 590 on Linux with working hardware acceleration but very slow performance...

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-RX-590-Linux-Workaround

[CochainComplex]

mhh...maybe I'm thinking not deep enough..but in a first approach RX590 seems to be an overclocked RX580. How can it make such difficulties to provide stable Linux experiance? Whats the culprit?

[dungeon]

mhh...maybe I'm thinking not deep enough..but in a first approach RX590 seems to be an overclocked RX580.

RX 590 isn't overclocked RX 580, it is entirely different asic

Various RX590 comes on a cards differently setuped and clocked than referent RX 590 one, and that is an issue here it seems

How can it make such difficulties to provide stable Linux experiance? Whats the culprit?

Stable VS Unstable Linux experience difference, could be even cos of one letter typo somewhere

[darkbasic]

LOL it's still enough to play those games at 4K.

[geearf]

Shouldn't the card use pp instead of dpm?

[CochainComplex]

RX 590 isn't overclocked RX 580, it is entirely different asic

Various RX590 comes on a cards differently setuped and clocked than referent RX 590 one, and that is an issue here it seems

Stable VS Unstable Linux experience difference, could be even cos of one letter typo somewhere

could be worse expensive/expansive ...About new ASIC ...mh not so sure - yes not just overclocked but according to this german article:

https://www.golem.de/news/radeon-rx-590-im-test-amds-grafikkarte-leistet-viel-fuer-ihren-preis-1811-137568.html

Polaris 30 has same transistorcount and the same die size only the structure was changed 14LPP -> 12LP compared to Polaris 20.
Yes it is new, but it is a "refresh" not entirely new.

So I would have expected less issues - since it is not a complete new plattform.

[dungeon]

Yes it is new, but it is a "refresh" not entirely new.

Well, nothing is entirely new

Whatever that have different pci id, is considered different.

As this card model is OC a bit more than referent RX 590 clocks, maybe it would help if Michael could patch kernel driver to force card to start at referent clocks

It is not first time that driver need temporal quirks like that

[CochainComplex]

Well, nothing is entirely new

Whatever that have different pci id, is considered different.

As this card is OC a bit more than referent RX 590 clocks, maybe it would help if Michael could patch kernel driver to force card to start at referent clocks

True but the gap is not like Polaris to Vega. .... He could also modifiy the vBIOS to set it to ref clock. Don't know if the tools to do this are allready supporting RX590.

[dungeon]

True but the gap is not like Polaris to Vega. .... He could also modifiy the vBIOS to set it to ref clock. Don't know if the tools to do this are allready supporting RX590.

I am talking that all chips are different and that it is not first time that dpm have stability problems... see a bug from years before, these cards also couldn't start with fab OC:

https://bugs.freedesktop.org/show_bug.cgi?id=76490

Sapphire Nitro r7 370 worked fine on FGLRX, but failed with radeon dpm with default clocks... and now another nitros, even that fatgirl

Whats the culprit?

Culprit is as people wanna hack & roll, meanwhile one harassment free letter typo somewhere is enough to break everything

[Xinkai]

The workaround works with my XFX card. Thanks to Nate for the discovery and sharing of this.