This may or may not be related, but wanted to warn you about possible differences in command lines.
Under Linux (or any *NIX with an X11 graphics server) there is a possibility of more than one GUI server. When that is the case the software must know which server to talk to. The answer is from the “DISPLAY” environment variable: If the person running the command has a DISPLAY set, and if that person has permission to use that DISPLAY, then this is where the command renders. There are other side-effects.
If you want to know the nuances of GPU remote commands versus local commands, when an X GUI server is involved (including GPU computations which are not graphical), then see:
https://forums.developer.nvidia.com/t/video-output-does-not-show-up-because-opengl-failed-to-create-x11-window-imagenet-camera-failed-to-create-opengl-display/68681/7
As a point of mentioning this, the library required, when associated with a GPU’s X server (whether used for display purposes or used for CUDA computations), must be provided by the system providing the display. Virtual servers are quite different than remote forwarded servers (virtual servers tend to do what you expect, remote forwarding won’t be what you expect in terms of required libraries).