Can I run a CUDA Program on a computer if I remote into it using Windows RDP? [duplicate]

Question

I have a server that is located elsewhere. On that server, I plan to run a program written to take advantage of the CUDA-capable graphics card on that physical machine. This program has nothing to do with graphics--it will only use the graphics card for computational math.

Will I be able to run this program on that machine? Will the program recognize the physical graphics card if I am logged in using windows remote desktop? Or will the server not even recognize that there is a graphics card plugged in since I am remoting in?

EDIT: I wanted to mention that all computation will take place remotely and that nothing on this machine depends on what's going on locally on my home PC. I am simply using my home PC to "view" what is going on at that server.

Answer 1

You cannot use Remote Desktop to run CUDA. Remote Desktop unloads the graphics driver which disables CUDA.

You need to use VNC or Splashtop or something else instead of Remote Desktop.

Answer 2

RDP runs in Session 0 -- just like a service. Graphics drivers aren't accessible from Session 0, so normally CUDA programs won't work (because they won't be able to access the underlying hardware).

If you're using a Tesla, you can put the driver into Tesla Compute Cluster (TCC) mode and bypass this issue, so any CUDA programs will work just fine over RDP.