This page answers common questions about accessing and using the POLARIS HPC cluster, including basic specifications, file transfer, Slurm commands, and GPU/CUDA checks.

General Information

Who can access the HPC cluster?

Access is typically granted to SFSU faculty, researchers, and sponsored students who require high-performance computing resources for research or curriculum-based projects.

How do I request access to the HPC cluster?

Submit a request via ServiceNOW or email at@sfsu.edu. Include a brief description of your project, whether you need CPU or GPU resources, and an estimate of your runtime and storage needs.

What are the specifications of the HPC cluster?

Network: InfiniBand (HPE HDR/Ethernet 200Gb).
Scheduler: Slurm.
GPU: NVIDIA A100 (80GB) available on GPU partitions.

Access and Usage

How do I log in to the HPC cluster?

Follow this guide: Accessing the HPC cluster .

What software is available on the HPC cluster?

Common software includes OpenMPI, Slurm, and CUDA (GPU partitions). Availability may change over time. To request additional software, submit a ServiceNOW request or email at@sfsu.edu.

How do I submit a job using Slurm?

See our guide: Job Submission using Slurm .

How do I transfer files to and from the HPC cluster?

Use scp (or rsync for large transfers). Examples:

Copy a file to the cluster:

scp /path/to/local_file your_username@HOSTNAME:/path/to/destination/

Copy a file from the cluster to your computer:

scp your_username@HOSTNAME:/path/to/remote_file /path/to/local_destination/

Windows users can use a file transfer client such as WinSCP.

Technical Details

How do I check what partitions are available and what their status is?

sinfo

What type of network is used in the HPC cluster?

The cluster uses InfiniBand networking (HPE HDR/Ethernet 200Gb switch).

What GPUs are available?

GPU partitions provide 4× NVIDIA A100 (80GB) per node.

How can I check the CUDA/driver version on the GPU nodes?

You can view GPU and driver details using:

nvidia-smi

Example (run on a GPU node via an interactive job):

salloc --partition=gpucluster --gres=gpu:1 --time=00:10:00
nvidia-smi

How can I use nvidia-smi in a script?

Example Slurm job script snippet:

#!/bin/bash
#SBATCH --partition=gpucluster
#SBATCH --gres=gpu:1
#SBATCH --time=00:05:00
#SBATCH --output=gpu_info.log

nvidia-smi

Troubleshooting and Support

Who do I contact for technical support?

Contact the Academic Technology Systems Team by submitting a ServiceNOW request or emailing at@sfsu.edu.

What should I include if I encounter an error while submitting a job?

• The job ID (from squeue or sacct)
• Your job script
• Your error/output logs (--output and --error)
• The exact command you ran
• Any relevant error message text

How do I check the status of my job?

Your jobs:

squeue -u your_username

Completed jobs:

sacct -u your_username

How do I cancel a job that I’ve submitted?

scancel jobID