Running CUDA Applications using Remote Desktop

We are trying to run a CUDA Application from a using remoter desktop.
The application works fine when we run it on the machine, but when we run it remotely, it gives the following error

Exception in thread “main” CUDA Driver error: 100
at jcuda.CUDA.setError(
at jcuda.CUDA.init(
at jcuda.CUDA.(
at cudatest1.LoadModule.main(
Java Result: 1

We are thinking that since two users cannot be active at a single time on the machine, the GPU becomes inactive when we login remotely. Is it correct, and if it is, then how to keep the GPU running?

Any help will be appreciated.
Please reply as soon as possible, we have to get it working before Monday.

Everytime I’ve tried running my CUDA application in a RemoteDesktop session the listing of CUDA devices fails for some reason. Even after logging out and getting on the system directly it still fails until a reboot. My solution for remotely using the CUDA board is to set up sshd with cygwin on the CUDA PC and ssh in from my main development machine (without CUDA). I am able to remotely run through the ssh session.


Remote Desktop replaces the normal video driver with a special driver to make capturing screen updates more efficient. This has the effect of making CUDA (which relies on the NVIDIA video driver) stop working when Remote Desktop is active. A solution is to use VNC, which does not replace the video driver.

Thanks a ton guys.
VNC works just fine.