Using only 1 graphics card for computation In which order are cards reported


I have built a CUDA application that will run on a pc with two graphics cards installed, which might be of the same type. I only want to use 1 of those cards for my CUDA computations. The other cards will be used for processing and displaying with DirectX, and will be connected to a monitor.

  • Is there a way to detect which card is used for display and which card I can use for CUDA?
  • Is there maybe some defined order in which the cards are enumerated if I call cudaGetDeviceProperties?

tnx, Arno

In reply to myself:

  • I have still not been able to detect which card is used for display
  • The cards are enumerated in a defined order: the card that is installed first will be listed first. But no conclusions can be made about whether the card is used by Windows (for display purposes).

Maybe cudaChooseDevice could be of any interest, but the documentation doesn’t give any information about which device properties (and in which order) are used for the choice of the graphics card.