Comment # 38 on bug 1173733 from Matthias Bach

(In reply to Stefan Dirsch from comment #29)
> > getfacl /dev/nvidia*
> > [...]
> @Mister Pend Seeems only gdm has access to your nvidia devices? It should
> look different once a regular user has logged into the session.

Just to make this explicit, it's completely valid to utilise the compute
capabilities (nvenc, CUDA, OpenCL) without a running X session, i.e. for a
dedicated machine-learning host that for noise reasons you don't want to have
right next to your desk. It's one of the big advantages of the NVIDIA cards
over AMD that you can have a truly headless system with them. The first
generation of Tesla cards didn't even have graphics outlets, and in consequence
wouldn't work on Windows (which I assume is the only reason why a lot of
supercomputers now could run giant display farms).