How to add second CUDA device to system? One for display, one for compute

I have my system (Windows 7) set up with one CUDA device, but it is unfortunately also the display card so I get visual artifacts and system crashes from time to time (plus I worry about hitting that 5-second kernel runtime limit). I have another CUDA device I want to insert in the system, and use one for compute and one for display.

The question is - how do I do that? I know how to physically insert the card, but I’ve never had two GPUs in a system before so I’m not sure how to configure the software. Do I reinstall the nVidia drivers and CUDA software kit? And how do I specify that one card is for display and the OTHER is for CUDA computation?

Thanks for your help!


Surely there’s a pretty standard procedure for this considering you have to do it if your want to make kernels run for more than 8 seconds… There’s a surprising lack of documentation or talk about this subject.