Multiple GPU

Hi,

It is disappointing to see that the 3.0 spec did not address multiple gpu. This mirrors a thread I started back in 3/13/08 OpenGL under windows forum (subject Multiple GPU).

Here’s a request that goes out to the Khronos group so they’ll consider adding multiple GPU in the future spec.

We can do this using Quadro card with WGL_NV_gpu_affinity. Can we please make this a standard?

Today, I am able to access multiple GPU using Direct-X and setting the cards in a multiple monitor setup. Why can’t we do the same with OpenGL.

Kovalan

You can with NVidia on Linux. Just create two screens, one per GPU.

You can do it without Nvidia, too, you just need to enabled Xinerama. With NVIDIA it’s better to use their proprietary system (I foget what it’s called at the moment).

Of course, there’s no easy way to know which GPU is on which screen. The Xlib functions to deal with screens are frightening to say the least.

You can do it without Nvidia, too, you just need to enabled Xinerama. With NVIDIA it’s better to use their proprietary system (I foget what it’s called at the moment).[/QUOTE]
TwinView.

Of course, there’s no easy way to know which GPU is on which screen.

Sure about that? You can specify Devices with PCI IDs, and Screens to use specific Devices. Precise control of which GPU is which screen IIRC. Also, with separate screens, there’s (according to NVidia) very little overhead for this. And no vaguery of which GPUs get which textures/etc. as with TwinView/Xinerama.

If you know how the xorg.conf is setup, that’s true. I was referring to the general case of having two screens set up, and trying to figure out though X calls what sort of GPU is bound to each screen before you create a context.

I guess I forget to specify that the problem is on a MS Windows platform. There are work around to get this working, but it should be part of the OpenGL spec.