7 GPU SLI??

[knaughty]

Oh, and if you're wondering WTF could use so much physical hardware, here's the answer:

OpenGL implemented in Silicon

[Kelaan]

I don't them to function together in SLI at all. Just be available for computation.

I'm pretty sure I read an article last year about someone using a many-GPU rig for CUDA. CUDA 4.0 is supposed to support using multiple cards in the same system:

http://developer.nvidia.com/cuda-toolkit-40

Share GPUs across multiple threads
Use all GPUs in the system concurrently from a single host thread

From what (VERY LITTLE) I've read in the past ten minutes, it sounds like current CUDA requires you to optimize/write code yourself which will accomodate multiple GPUs, if it's even supported, but that the Fermi architecture will (or does? is it out already?) should make that more transparent.

Out of curiosity, what kind of processing are you looking to do that would benefit from that many GPUs? Have you already made it work with a single card? Some of the current cards already have several hundred CUDA cores in them, and might be sufficient for your needs, if you haven't already determined otherwise.


edit: Well, that's what I get for not reading the whole thread before replying. Glad you have a clearer idea of what you need. Also, holy crap those rendered pictures look awesome. I wonder if the rendering process is parallelizable across machines, so that you could render one frame (or part of an image?) on one machine and another on a second machine?