Slurm Quick Start

Most users only need four steps to get started:

1. Choose a partition
2. Write a batch script
3. Submit it withsbatch
4. Monitor it with squeue or sacct

Minimal Batch Script

Create a file named hello_slurm.sh:

#!/bin/bash -l
#SBATCH --partition=scu-cpu
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=1
#SBATCH --job-name=hello_slurm
#SBATCH --time=00:02:00
#SBATCH --mem=1G
#SBATCH --output=hello_slurm-%j.out
#SBATCH --error=hello_slurm-%j.err

source ~/.bashrc

echo "Starting at: $(date)"
echo "Job ID: $SLURM_JOB_ID"
echo "Node: $(hostname)"
echo "Cluster: $SLURM_CLUSTER_NAME"
echo "TMPDIR: $TMPDIR"
sleep 30

Submit the Job

sbatch hello_slurm.sh

Slurm will return a job ID. By default, Slurm writes output to a file such as slurm-<jobid>.out unless you override it with --output.

Check Job Status

squeue -u <cwid>

Check Completed Jobs

sacct -u <cwid> -S $(date -d "-7 days" +%D)

Quick Interactive Session

srun -n1 --pty --partition=scu-cpu --mem=8G bash -i

Quick GPU Interactive Session

srun -n1 --pty --partition=scu-gpu --mem=16G --gres=gpu:1 bash -i

GPU requests are often site-configured as GRES, so --gres=gpu:1 is a common pattern.