[NLPL Task Force (A)] [rt.uio.no #3552285] gpu usage on Abel for teaching in october

[Stephan Oepen via RT]

many thanks for the quick follow-up, sabry!

> May I recommend to use one of the ML machines for this ?

truth be told, i would like to search for a more 'official' solution.
the teaching in this case will actually be done by our swedish
collaborators (from uppsala, copied), so getting access for everyone
to the ML nodes could be a first obstacle.  also, the necessary
software is not trivial to install (OpenNMT on top of PyTorch, plus a
few pre-processing packages), whereas the NLPL project directory on
Abel already provides both the software and data.  finally, we
actually want the students to experience the cluster (and learn how to
properly request allocations and submit jobs), because at least some
of them will move on to become cluster users for their MSc thesis
projects.

regarding Abel availability during the course duration, that is of
course an important consideration too.  given past experience,
however, i might worry more about a large enough number of gpu users
being migrated away over the next three weeks already than about Abel
actually going out of service at the end of october: i think the most
recent HPC newsletter indicated a final target date 'by the end of
2019', though was somewhat non-committal about that :-).

good night for today :-), oe