Will matlab use CUDA acceleration on a hybrid AMD-nVidia system (Radeon as output device)

Hello everyone.

I am planning to put together a system for both gaming and computing. I got interested in nVidia's CUDA technology and among other features - it's potential to accelerate Matlab computations. I haven't used CUDA before at all.

Now I'm thinking of using a decent Radeon video card as a primary videocard and a budget nVidia card as a dedicated PhysX (for games) and CUDA (for computing) card. Now, I lurked the Net and it looks like this setup can work (albeit with some installation issues) and CUDA will work as long as there is no need to make a direct output from the nVidia card.

THE QUESTION is - will Matlab recognize and use the nVidia card's resources in such a setup?

I haven't decided on the exact cards yet, but I'm guessing it will be something high-end from the previous generation of Radeon card and something mid-range from a generation or two back for the nVidia card, supporting PhysX and CUDA.

Thank you.

Answer 1

[Jason Ross]

The thing that matters is the compute capability of the CUDA card. It needs to be version 1.3 or higher. There is a nice table of the cards and their compute capabilities here:

http://developer.nvidia.com/cuda-gpus

This setup should work on Windows 7. I recall digging through the install instructions for CUDA and seeing that it won't work on Windows XP, but 7 should be fine.

Also, keep in mind that the entry-level CUDA cards are not very powerful in terms of acceleration. You can still program against them, but the results are likely not to be that impressive.

In my personal opinion, I like to try and keep things as simple as possible and would just find two nVidia cards that would work, using one for display and the other as computation.