There Are A Few More Performance Changes With RadeonSI From Mesa Git

[phoronix]

Phoronix: There Are A Few More Performance Changes With RadeonSI From Mesa Git

With Marek's optimizations having landed in Mesa Git that targeted Deus Ex: Mankind Divided, I ran benchmarks and found Deus Ex: MD is generally much faster and can be 2~3x faster, much more than the 70% originally thought by Marek. Now that more time has passed, I have carried out some more Linux gaming tests...

http://www.phoronix.com/scan.php?page=news_item&px=RadeonSI-Git-More-Than-Deus

[Veerappan]

Metro Last Light Redux saw a boost for the R9 Fury at 4K, but at 1080p the three cards tested saw their performance unchanged. The R9 285 at 4K with metroll-redux appears to have regressed quite a bit.

Unless I'm reading that chart wrong, the Fury regressed pretty hard in that test, not improved... But it is still.pre-coffee here.

[Michael]

Unless I'm reading that chart wrong, the Fury regressed pretty hard in that test, not improved... But it is still.pre-coffee here.

Whoops, yep, fixed that sentence. Not enough coffee here yet :P

[LeJimster]

Overall very impressive. I'm hoping to see more of this before Vega hits. Can't wait.

[MadCatX]

I wonder if those regressions have anything to do with Marek's patches or if it's a result of some unrelated change that made it recently to mesa master. Anyway, the potential of Marek's patches is obvious and pretty exciting!

[Azpegath]

Wow, this is amazing to see, finally the Fury is performing more in line with its price!

[Jumbotron]

This is why open source is like the story of the rabbit and the turtle. Nvidia and their proprietary rabbit drivers may be faster now.....but open source keeps chugging along and eventually that rabbit is going to be VERY surprised when they see the turtle right beside them saying...."Hey....what's up?"

Once again...hat's off to Marek and all the others working on open source GPU drivers of ALL makes and models.

[helland]

Michael Do you have an openbenchmarking url for this? I'm trying to bisect this, to find the root of the Metro LL regression. However, I can't seem to be able to reproduce the results from your December 22th article with my RX460 by reverting my mesa install. So it might instead be kernel or llvm related issues. But to find out what to look at next, it would be nice to know if the kernel or llvm was similar to your last run in December.

[SomeoneElse]

Michael Do you have an openbenchmarking url for this? I'm trying to bisect this, to find the root of the Metro LL regression. However, I can't seem to be able to reproduce the results from your December 22th article with my RX460 by reverting my mesa install. So it might instead be kernel or llvm related issues. But to find out what to look at next, it would be nice to know if the kernel or llvm was similar to your last run in December.

In the bottom left corner of each chart, there is a blue cube. Click it an you will be redirected to the corresponding openbenchmarking.org results. At the end of the result page there is also the link to the overview with all results.

[geearf]

This is why open source is like the story of the rabbit and the turtle.

It's not a rabbit and a turtle but a hare and a tortoise