[NLPL Task Force (A)] [rt.uio.no #3396801] newer CUDA versions on Abel

[Stephan Oepen via RT]

> Could you try to use Anaconda and create an environment with Python3.6 as base? I've done this on the ml servers. No problem. Of course, I'm not sure if Python3.6 is all you need.

yes, that would probably work (once i work out how to best wrap an
anaconda environment as a module definition).  or i could probably
compile my own Python 3.6, where i would have to make sure to pick up
the right set of libraries (e.g. a newer version of SSL than the
default OS one on Abel).  ideally, i would then also work out how to
compile my own NumPy add-on, using MKL as the back-end ...

but all this would feel like stepping into USIT territory, i.e. so far
i have been working on the assumption that Python and CUDA are among
those core components that are uniform for all users, not maintained
(and hidden) in a discipline-specific project.

so, is it actually not so easy to just provision the CUDA 10 libraries
as another module, without touching the drivers?  or are we running up
against a matter of principle here, no more software updates on Abel?

cheers, oe