sheffield_hpc
sheffield_hpc copied to clipboard
rcgsheffield
Docs for University of Sheffield HPC systems
>We may want to generate a short importable section with the crib notes of how to check this for dependency heavy software / modules like this. https://github.com/rcgsheffield/sheffield_hpc/pull/1775#discussion_r1183660354 This may be...
Current list: - NetLogo - https://github.com/rcgsheffield/sheffield_hpc/pull/1755
The diagram here: https://slurm.schedmd.com/mc_support.html#defs  Is likely worth an add to our reference info section where it discusses, CPUs, cores and threads.
[Installing software to the clusters](https://docs.hpc.shef.ac.uk/en/latest/hpc/installing-software.html) (Non-essential) - Easybuild for Bessemer / Stanage will need to be added eventually. (Less deconfliction than tabs + deduplication of effort.)
Can we throw together some info about the A100 minimum CUDA versions or a sensible recommended minimum which may just be the 11.X on system at present.
As per a bit of discussion here. We should revisit this in due course: https://github.com/rcgsheffield/sheffield_hpc/pull/1688#discussion_r1158608660
This should be using SLURM native tooling (srun) ideally. Where we have the choice of mpirun in srun, the latter is preferable but we may want to discuss/document why. Clarifying...
At the moment the initial paragraph makes it sound like you get access via purchasing a node or by being part of a research group that has done. We should...
Similar to [#using-conda-and-python-in-a-batch-job](https://docs.hpc.shef.ac.uk/en/latest/bessemer/software/apps/python.html#using-conda-and-python-in-a-batch-job)
As it stands, the Bessemer documentation looks coherent on coupling with the toolchain GCC versioning so there's nothing to fix there. That said I think a compromise here would be...