JupyterHub¶
CARC runs JupyterHub portals that launch Jupyter notebook servers on cluster compute nodes, so your notebooks have direct access to CARC storage, modules, and hardware.
| Cluster | JupyterHub URL |
|---|---|
| Hopper | hopper.alliance.unm.edu |
| Easley | easley.alliance.unm.edu/jupyter |
Log in with your CARC username and password. New here? Start with Getting started at CARC.
Starting a server¶
After login, JupyterHub asks for your session options (Slurm account, resources, and duration — the exact form depends on the cluster). Your notebook server is a Slurm job under the hood, so:
- it may queue briefly until resources are available;
- it stops when its time limit is reached — save your notebooks;
- the resources you hold count against your project's fairshare, so stop your server (File → Hub Control Panel → Stop My Server) when you are done.
Using your own environments as kernels¶
The default kernels cover common cases, but most research needs its own packages. Create a conda environment and register it as a Jupyter kernel — the walkthrough is in Conda environments in JupyterHub.
Scaling beyond one node¶
Notebooks don't have to stay single-threaded:
- Parallel Python with Dask and scikit-learn - scale scikit-learn across cluster workers from a notebook.
- MPI parallelization from JupyterHub - drive mpi4py/ipyparallel from a notebook session.
- For long or heavy computations, move to a batch job — notebooks are for exploration, batch is for production runs.
Related pages¶
- Open OnDemand - browser access to files, shells, and other interactive apps.
- Installing deep learning packages - GPU-ready PyTorch/TensorFlow environments.
- Getting help - stuck server or missing kernel? Open a ticket.