Is the onboard GPU a CUDA device? If not, just connect your monitor to the onboard device, don’t extend the GPU onto the Quadro, and run CUDA normally. Otherwise, you’ll have to use cudaSelectDevice().
Simliar problem I have met. It seems that the issue depends on the MotherBoard.
I have a card Quadro FX5600 with an IGD (GMA X3000) on board SuperMicro C2SBA. I plan to choose the GMX 3000 as the Display device and the FX5600 as the CUDA compute device. But they cann’t see each other in Windows XP. When I choose the IGD as the Default Primary VGA, FX5600 is disappeared in Device Manager and GMA X3000 works fine. When I choose the PEG, PCI or Auto, FX5600 is choosed and GMA X3000 is disappeared. It seems that they cannot coexist with each other.
By the way, if you find the solution, could you send me a mail? Also I will if I do.
Generally, integrated and discreet graphics solutions will not work at the same time. If you need a dedicated CUDA compute device, check out the Tesla C1060. http://www.nvidia.com/object/tesla_c1060.html Since it’s not a display device, it shouldn’t have any compatibility issues with an integrated graphics solution.