Radeon Graphics Yield Mixed Results With Linux 3.17 Kernel

[phoronix]

Phoronix: Radeon Graphics Yield Mixed Results With Linux 3.17 Kernel

Earlier this week I ran the article about open-source Radeon graphics having some improvements with Linux 3.17 but after further testing of a range of AMD GPUs spanning multiple generations, for not all hardware are AMD graphics cards faster with this experimental kernel. In this article are benchmarks of the Linux 3.17 kernel with Mesa 10.3-devel showing the winners and losers from the Radeon HD 5000 series through the latest Radeon Rx 200 series graphics cards.

http://www.phoronix.com/vr.php?view=20799

[schmidtbag]

As an owner of a HD5750, looks like I'll have dodged a bullet there. I'd rather there be no performance change at all than a regression. It's too bad though, I was hoping for an improvement.

[Dukenukemx]

As an owner of a HD5750, looks like I'll have dodged a bullet there. I'd rather there be no performance change at all than a regression. It's too bad though, I was hoping for an improvement.

I believe a 5750 and a 6750 are identical in hardware. The 6870 is different I believe. BTW, don't throw away those old Radeons just cause it's considered old. They're very comparable to modern SI cards.

http://hexus.net/tech/reviews/graphics/36269-amd-radeon-hd-7850-vs-6850-vs-5850-clocks/

[curaga]

Here's hoping Michael will bisect the 6870 after a week of pleading

[Creak]

Mesa vs OpenGL matrix

Hi there!

I made a PHP script that tries to display the OpenGL compliance of Mesa across the different drivers:

http://mesamatrix.net

It's on GPLv3.
It's not perfect yet, so if you see some bugs, feel free to tell me (github would be the best for that: https://github.com/MightyCreak/mesamatrix)

Cheers!

[tomtomme]

@Michael: does this make sense? At 20 august your benchmark of r270x showed improvements for. On kernel3.17 everywhere. Those are completely gone now... Did something change in your setup or in the kernel since then?

[My8th]

Hi there!

I made a PHP script that tries to display the OpenGL compliance of Mesa across the different drivers:

http://creak.foolstep.com/mesamatrix/

It's on GPLv3.
It's not perfect yet, so if you see some bugs, feel free to tell me (github would be the best for that: https://github.com/MightyCreak/mesamatrix)

Cheers!

Well done, much easier view showing how well Mesa is doing.

[tomtomme]

Hi there!

I made a PHP script that tries to display the OpenGL compliance of Mesa across the different drivers:

http://creak.foolstep.com/mesamatrix/

It's on GPLv3.
It's not perfect yet, so if you see some bugs, feel free to tell me (github would be the best for that: https://github.com/MightyCreak/mesamatrix)

Cheers!

Can you explain the matrix a bit more?
I thought we were pretty close to openGL 4.0 for intel, but it looks like this is only true for mesa but not for the real intel driver or the other drivers. Does every mesa feature have to be implemented in extra in every driver? how much work is this?

[Isedonde]

I thought we were pretty close to openGL 4.0 for intel, but it looks like this is only true for mesa but not for the real intel driver or the other drivers.

Just three GL 4 features (out of 25) are missing for the i965 driver, so that's pretty close AFAICT. Of course, tessellation might not be very easy to implement, so even though that is only 1 feature it might be more work than all the other features combined, I don't know.

@Creak: Very well done ? This is much easier to read than the GL3.txt. I was actually thinking about doing the same, but was too lazy to do it. :-) Thanks! I know there's some other matrix view, but that was very outdated last time I checked. So please keep that script on your host (or try to convince mesa about adding this to their web site if you can't host it for the foreseeable future :-) ).


@Topic: As a 6870 owner, this makes me sad. I hope someone with spare computing time and a 6870 can bisect this.

[dungeon]

I think those are change, so how vbo is handled:

http://cgit.freedesktop.org/mesa/mesa/commit/?id=8898fff46cbf62baf4f924b6e8f537a9b1b26fa5

http://cgit.freedesktop.org/mesa/mesa/commit/?id=150ac07b855b5c5f879bf6ce9ca421ccd1a6c938

So revert second patch to bring back older behavior or play with the values like in bug https://bugs.freedesktop.org/show_bug.cgi?id=82050

Not sure it will stay like that because there is stutter in some apps, for example in Unigine Valley .

Other way around (6950 excluded) hardware now play more in line with expectation