HPC Learning Hub is a simulated cluster that runs in your browser. It exists because the hardest part of learning HPC is getting access to a cluster in the first place, and because the second hardest part is being willing to break one.
It is free, needs no account, and installs nothing.
A login node and twenty-one compute nodes across six partitions, with a scheduler
that genuinely allocates them. Submitting a job takes cores from named machines;
cancelling it gives them back. squeue, sinfo and
scontrol show node are computed from one record of who holds what, so
they cannot contradict each other. Actions are written to the daemon logs, in the
format real Slurm uses, and you can grep for your own job.
Underneath is a Linux shell: pipes, redirects, command substitution, separate stdout and stderr, vim and nano, ssh between nodes, a filesystem with permissions and ownership. That part is shared with KubernetesTrainer, where it has had far more use than anything here has yet.
Real compilation. A build takes seconds here and hours in reality, and the errors you meet are the common ones, not yours. Real MPI, real numerics, real IO contention, real filesystem behaviour under load. Anything involving the actual speed or output of scientific software.
Where a command accepts an option that has no effect, the site will tell you so rather than let you assume it worked. That is a deliberate feature, not a limitation being confessed.
People about to get an account on a real cluster and hoping not to waste the first week. Researchers who inherited someone else's batch scripts. Anyone who has been told to "just submit it with sbatch" and would rather not learn by failing in public.
It is not a certification course. HPC does not have one.
Three separate sites — LearnSlurm, LearnSpack and LearnApptainer — each with its own single-machine terminal. On a real cluster you do not meet these tools separately, so they have been merged onto one set of machines. The old addresses lead here.
← Back to the cluster