Using CUDA through VPN

Hi all ;)

I am doing a CUDA project alongisde a colleague of mine, althought the projects are somewhat independent. We only have one PC(Linux) with a CUDA compatible card at work, so we started using VPN.
Recently we were having strange problems when trying to compile programs in our Windows computer connected though Secure Shell Client and VNCViewer to the Linux computer. When we copied the files to the Linux computer the files were compiled and ran right, so given that I was always compiling my files in the Linux computer without major problems, I tried to run them through VPN and got errors from all the CudaMallocs, CudaMemcpy,CudaFree etc.

Why does it happen? One of the programs was the sum of a vector with 10 elements using thrust and it returned error std::bac_alloc(which happens to allocations that surpass the card maximum memory)

Thanks in advance

Are you sure that both of you have write privileges to the /dev/nvidia* files?

No sorry, I do not know what files are those. Anyway why should we not have priviliges only when we are connected through VPN?

It started working for both of us out of nowhere, very strange but good news nevertheless