Quick Start Guide

Get up and running on the cluster in about 15 minutes.

Tip

This is the express lane. For the long-form version see Getting Started. For the full reference, follow links in each section.

1. Get an account 

Before you can log in, three things must happen — the first is the slowest, so start now:

Your PI emails help-hpc@caltech.edu asking to add you to their group. (New groups: see Account Information.)
Set up Multi-Factor Authentication at access.caltech.edu/my_duo. Duo Mobile on your phone is the easiest option.
Complete eligibility certification at access.caltech.edu/hpc_portal. This is the export-control acknowledgment — see Policies.

You’ll get a confirmation email when your account is ready.

You must be on the Caltech network — either physically on-campus or connected to Caltech VPN. SSH attempts from anywhere else will hang or be refused. See Common Problems → Connection Refused if the login host is unreachable.

macOS / Linux

Open a terminal and connect:

ssh username@login.hpc.caltech.edu

You’ll be prompted for your access.caltech password and a Duo push.

Windows 10/11

Use the built-in OpenSSH client from PowerShell or Windows Terminal:

ssh username@login.hpc.caltech.edu

Or use MobaXterm for a graphical SSH/SFTP client.

Web (no install)

Use Open OnDemand — a browser-based interface for shells, file management, and interactive apps (Jupyter, RStudio, MATLAB).

Important

You must SSH at least once before using Open OnDemand — the first SSH login creates your home directory.

Recommended SSH config 

Save typing and avoid re-authenticating constantly. Add to ~/.ssh/config on your local machine:

Host hpc
    HostName login.hpc.caltech.edu
    User YOUR_USERNAME
    ServerAliveInterval 60
    ControlMaster auto
    ControlPath ~/.ssh/sockets/%r@%h:%p
    ControlPersist 10m

Then create the sockets directory once: mkdir -p ~/.ssh/sockets. After this, ssh hpc is enough.

3. Move some data over 

# Upload a single file
scp myfile.txt hpc:~/

# Upload a directory
scp -r mydata/ hpc:/resnick/groups/yourgroup/$USER/

# Download results
scp hpc:/resnick/scratch/$USER/results.txt ./

For large transfers (> ~100 GB), use Globus instead — see Transferring Files.

4. Find and load software 

module avail              # List everything available
module spider pytorch     # Search for a specific package
module load python3/3.10.12
module list               # Show what's loaded

If something’s missing, request it via help-hpc@caltech.edu — or install it yourself with conda or pip.

5. Submit your first job 

Save the following as hello.sh:

#!/bin/bash
#SBATCH --job-name=hello
#SBATCH --output=hello-%j.out
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=2
#SBATCH --mem=4G
#SBATCH --time=00:05:00

echo "Hello from $(hostname) at $(date)"
echo "Allocated $SLURM_CPUS_PER_TASK CPUs and $SLURM_MEM_PER_NODE MB RAM"
sleep 30
echo "Done."

Submit, watch, and inspect:

sbatch hello.sh           # → "Submitted batch job 12345"
squeue -u $USER           # Watch it run
cat hello-12345.out       # See the output
seff 12345                # Check how efficient the request was

Tip

seff tells you how much of your requested CPU/memory you actually used. Right-sizing your next job by that ratio is the single biggest thing you can do to improve queue times.

6. Cheat sheet 

Command	What it does
`sbatch script.sh`	Submit a batch job
`squeue -u $USER`	Show your jobs
`scancel JOBID`	Cancel a job
`scancel -u $USER`	Cancel all your jobs
`srun --pty bash`	Quick interactive shell on a compute node
`salloc -N 1 -t 1:00:00`	Reserve resources for an interactive session
`seff JOBID`	Efficiency report after a job finishes
`sacct -j JOBID`	Detailed job accounting
`module avail`	List installed software
`hpcquota`	Check storage usage
`sinfo -p gpu`	Show GPU partition state

7. Where to put your files 

Location	Quota	Backed up?	Use for
`/home/$USER`	50 GB	No	Scripts, configs, dotfiles
`/resnick/groups/<group>`	20 TB	No	Project data, results to keep
`/resnick/scratch`	Large, shared	No	Active computations, temp files

Warning

Nothing on the cluster is backed up. Files on /resnick/scratch are purged after 14 days without access. Move important results to group storage and copy critical data offsite — see Backups.

Next steps 

Example Job Scripts

Copy-paste templates: serial, MPI, GPU, arrays, MATLAB, R, GROMACS, AlphaFold

Example Job Scripts

GPU Computing

H100/H200, CUDA, deep learning

GPU Computing

AI/ML Guide

PyTorch, TensorFlow, LLMs, distributed training

AI & Machine Learning

Best Practices

Right-sizing requests, checkpointing, I/O patterns

Best Practices

Stuck?

Troubleshooting — symptom-based index
Common Problems — fixes for the usual culprits
FAQ
Submit a ticket: Caltech Help System
Email: help-hpc@caltech.edu