Problems with multi-gpu settings

Dear All,

I have a gtx295 card and Vista x64. My problem is, that if the multi-gpu is disabled in the NVIDIA control panel, than CUDA sees only one device, and my program is working fine. But if I switch it on, than the program calculates stupid thing (although it does not use the second GPU).

        Does anybody has any idea why it could be?

One additional information. There is an iterative method (using double precision) in the program. If multi-GPU support is on, than in the iterative method random mistakes happen.