using one GPU for computations and display

Will I get any speedup after installing second graphic card for display?

I’d say only if your application is already graphics intensive.

AFAIK, if you have 2 GPUs you could perform graphics computations in parallel with CUDA computations.

If you’re not even rendering anything, I would not expect any performance difference. Unless you break your problem in two and use the second GPU to CUDA as well.

But I’m only guessing here. :)

The second card would be some older nVidia GPU, I don’t even remember exactly wat it was. Anyway it’s useless for CUDA purposes. I thought that with One GPU some processors are used for CUDA computations but some of them have to be used to refresh display which is 1200x1600 (pivot). Am I right or am I missing something?

Well if you have two GPUs and you use one for only GPGPU then this can use all of its computation power. But if you use only one GPU and still want some output on screen the GPU is also used for this purpose. So you have less memory available on the GPU for GPGPU. And about the time that is taken to refresh the screen I don’t know anything about that. But keep in mind that if you have only one GPU and you are using windows or X windows that there is a watchdog which kills the Kernel if the computation time is more than 5 seconds.

5 sec. limit is rather not a problem in my case. At least not now. The main reason for installing second GPU is to avoid hard reset after each out of memory read. Anyhow thank You both for Your posts :)

It could make a difference anyhow, when you can use some older card for the operating system GUI and rendering. Even the 2D rendering could take cycles away from your CUDA work.

The only problem that can arise is if there is a lot of traffic between the card that renders and the card that displays (using OpenGL interoperability). In this case, it might actually be slower to use two cards…

Thank You wumpus,
That’s a good point to remember, yet I’m creating a kind of graphics library without using OpenGL interop. so it shouldn’t be a problem. I’m not installing the second card yet, but after (if at all :) ) I do, I will put the conclusions in this topic :).

I have a machine with a non-nvidia workstation class card. I have an 8800GT on order to do CUDA with. Can I leave my non-nvidia card in for graphics and just add the 8800GT for CUDA?