Archives Discussions

[ljbade]

I have found a serious bug when using OpenGL with Catalyst 12.10 and all 12.11 Beta versions with a laptop that has "Radeon Dual Graphics" (ie APU + GPU) and "Switchable Graphics".

Basically if you program runs in High Performance mode then you get some sort of strange moire pattern on the framebuffer, likely some CrossFire artefact. If you set it to Power Saving the problem disappears.

It affects not only my program I am developing but also other OpenGL programs such as Geeks3D.com GPU Caps Viewer.

For more info see my post on the OpenGL forum: http://www.opengl.org/discussion_boards/showthread.php/179649-Strange-GL-frame-buffer-corruption?p=1...

Screenshot: http://i.imgur.com/0Nk9D.png

For some reason it does not affect some games such as Rage or Doom 3 BFG, which may be to do with some sort of AMD customised CrossFire profile for these games.

This bug is still in 12.11 Beta8 (OpenGL build 11991) which was released today, so I hope it gets fixed before the WHQL version comes out.

[gsellers]

Hi,

Thanks for the report. I doubt this is crossfire related as we generally don't support crossfire on switchblade systems, and even then, we white-list applications and so your own application wouldn't have a crossfire profile by default. It is likely an issue with transferring the final image from the discrete GPU where it was rendered to the integrated GPU which controls the display.

We'll try to reproduce it at our end and look for a fix.

Thanks,

Graham

[ljbade]

Interesting, I always thought my laptop used both GPU's when running a game, as tools like MSI Afterburner would indicate a load on both GPUs.

So High Performance mode renderers the framebuffer on the discrete GPU then copies it to the integrated GPU for display, and Low Power renders it on the integrated GPU directly.

Is that correct?

[gsellers]

Yes, that is correct. The reason that you see load on the integrated GPU is because it is driving the display and compositing the desktop. The application itself is rendered using the discrete GPU.

Cheers,

Graham