Multi-GPU configurations with CUDA 2.1

Accelerated Computing CUDA CUDA Programming and Performance
1

I have a configuration with one FXQuadro 5600 and two Teslas C870.

Using CUDA 2.0 and Windows XP, I am able to see the three cards.

Using CUDA 2.1 beta, Windows Vista and the same hardware configuration, I can only see the cards which are enabled for PhysX. In the nVidia Control Panel, I am offered the choice of enabling the two Teslas or the FXQuadro. I cannot enable all three cards.

In the CUDA release note 2.1 instead I find the following statement as a Vista specific issue:

“In order to run CUDA on a non-TESLA GPU, either the Windows desktop must be extended onto the GPU, or the GPU must be selected as the PhysX GPU.”

I am using the FX Quadro to run my monitor, however this card is not visible from CUDA unless I enable it for PhysX. But if I do that, then the two Teslas stop being visible.

Am I misunderstanding the documentation or am I struggling with an issue that will be resolved in the final 2.1 release?

Thanks for the help.

2

Jusss my 2 cents:

There is a refernce in CUDA manual that multi-GPU support will be supported only for GPUs of same type… Now what “type” means is not documented. I had earlier asked in the forums and thre was no proper answer for it. May b, it was there with future in mind…

Anyway, only an NVIDIA person can answer your specific problem (relating to physx…)