Since r6082, we no longer depended on libnvidia-encode directly (we use dlopen), and as of r6336 we can also build RPMs against CUDA 6, and fail to load the nvenc module at runtime, which makes it a lot easier to deploy nvenc.
This is done using autoreqprov AutoReq: no to disable automatic processing of requirements so yum won't complain about the missing libcuda.so dependency when we try to install, since CUDA is actually optional and will not be available on most installations.
Alternatively, we could use a hack like this one: Can some specific autodetected dependency be ignored upon rpmbuild.
All done already - this was just a tracker ticket.
r6349 takes this further: we no longer link against libcuda either, so r6350 re-enables the automatic dependency tracking in the RPM build which had been disabled in r6336. (don't know why I didn't think of this earlier: cleaner, safer, better!)
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/564