1.1. New User Checklist#

Use this checklist when onboarding to the Kempner AI cluster for the first time. It is intended as a step-by-step path through the main setup tasks, with links to the handbook sections that explain each topic in more detail.

1.1.1. Before Requesting Access#

Confirm that your work is eligible for Kempner AI cluster access. See Overview of Cluster.
Review the expected use cases, fair-use expectations, GPU-only policy, and communication norms in Cluster Usage Policies.

1.1.2. Request Your FASRC and Kempner AI Cluster Accounts#

1.1.2.1. FASRC Account#

If you already have a FASRC account set up, you may jump to the Kempner AI Cluster Account section below.

Request a FASRC account through the FASRC account request portal.
Select the correct PI and approver for your role. See Introduction and Basics.
Wait for your approver to approve the account request in the FASRC portal.
Set your FASRC account password when your account is created.

1.1.2.2. Kempner AI Cluster Account#

Once you have an active FASRC account, you can request access to the Kempner AI cluster:

Reach out to your PI and fill out the Kempner AI cluster account request form. Your PI will guide you through the form. See Introduction and Basics for whom to contact.
Allow up to two business days for the PI to confirm approval and for the account to be set up on the Kempner AI cluster.
If you have not heard back, email rchelp@rc.fas.harvard.edu with the subject kempner cluster account.

1.1.3. Set Up Required Authentication#

Install and configure OpenAuth two-factor authentication. See Installing OpenAuth 2FA.
Install the Cisco Secure Client (formerly AnyConnect) if you plan to use Open OnDemand or other VPN-only FASRC services.
Configure the FASRC VPN using vpn.rc.fas.harvard.edu. See Accessing the Cluster.
Confirm that you can generate an OpenAuth code from your phone, password manager, Duo Mobile, or Java desktop app.

1.1.4. Log In for the First Time#

Connect by SSH:

ssh <username>@login.rc.fas.harvard.edu

See SSH Access.

Confirm that you can log in with your FASRC password and OpenAuth code.
Remember that login nodes are for file management, job submission, and lightweight tasks only. Do not run compute-heavy code on login nodes.
If you prefer a browser interface, connect to the VPN and open Open OnDemand. See Open OnDemand and Open OnDemand.

1.1.5. Learn Where Files Should Go#

Find your home directory (/n/home<NN>/<username>) and understand its 100 GB persistent storage limit. Check current usage with df -h ~/. See Storage Options.
Find your lab directory at /n/holylabs/LABS/<your_lab_name> for persistent lab storage (4 TB per lab). See Storage Options.
Find your scratch directory under $SCRATCH/<your_lab_name> (typically /n/netscratch/<your_lab_name>) for active high-performance work (50 TB per lab), and review the 90-day scratch retention policy. See Storage Options.
Choose an appropriate data transfer method before moving files: scp or rsync for smaller transfers, and Globus for large transfers. See Data Transfer.

1.1.6. Set Up a Working Environment#

Learn how to inspect and load software modules with module avail, module load, module list, and module purge. See Software Modules.
Configure conda to use conda-forge as the default channel. See Setting conda-forge as the default channel.
Create a project-specific conda environment. See Creating a Conda Environment.
If you will use Jupyter or JupyterLab, install ipykernel in your conda environment. See Using a Conda Environment with Jupyter.
If you will use VSCode, review the remote development workflow in VSCode for Remote Dev.

1.1.7. Run a First GPU Job#

Learn the basics of SLURM partitions, accounts, and job submission. The Kempner partitions are GPU-only, so every job must request a GPU with --gres=gpu:. See Understanding SLURM.
Identify which SLURM (fairshare) account you should charge jobs to. If you are in multiple groups, confirm the right account before submitting jobs.
Start with a small interactive GPU allocation to test your environment, then connect to the allocated node. See Job Submission Basics.
```
salloc --partition=kempner --account=<your_account> --time=0-01:00 --mem=64G --gres=gpu:1 --cpus-per-task=16
```
Submit a small batch job with sbatch after your interactive test works. See Batch Jobs.
Check that your job requests a GPU and uses Kempner resources responsibly. Revisit Cluster Usage Policies before scaling up.

1.1.8. Get Help and Stay Connected#

Join the #cluster-users channel in the Kempner Slack space.
Use the Kempner Slack channel for Kempner-specific workflow questions, community advice, and handbook update suggestions.
Review FASRC training and support options in Support and Troubleshooting.
Direct your questions to rchelp@rc.fas.harvard.edu with the subject line containing the word kempner, such as kempner account setup.

New User Checklist

Contents