I don’t know if this is a bug or just a setup issue. Mostly one linux box should be able to use “ssh -Y” to access a remote linux box, and then run any program. If that program is command line, it would work normally…if the program is X11 GUI, it should run on the foreign host and display on the local linux machine. For example, typing “xterm” on an ssh -Y connection should bring up xterm running remotely but displaying locally. CUDA breaks this (I’m using CUDA 6.0 from Jetson with L4T R19.3, remote display on Fedora 19 using nvidia drivers, but no module nvidia).
What I’m seeing is failure due to needing module “nvidia” on the local machine. However, the CUDA program runs on the remote machine, and only the remote machine nvidia module should be required. If the local machine were any flavor of UNIX, running X11, it should just work without the display machine having module nvidia.
So is this a configuration issue, or a bug when it tries to use the nvidia module on the local display while running CUDA on a remote machine?