Open OnDemand is a web-based interface that provides access to HPC cluster resources through a browser. It allows you to launch interactive desktop sessions, run applications, manage files, and submit jobs without using the command line. Through Open OnDemand users can run an interactive desktop session on the GPU node of HPC2.

• Log in at https://hpc2.grinnell.edu using your Grinnell College credentials.
• Click the GPU Desktop tile under the *Interactive Apps* heading.

You will be presented with a job submission form. Since this interactive job is already configured, you'll only need to specify the amount of time (in hours) you expect the job to run. Click 'Launch' to submit the job to the scheduler. Your job will be placed in the queue.

You'll be directed to a page displaying your active and recent sessions. Your new session status will typically show:

• Queued
• Starting
• Running

Once the job starts, a button titled *Launch GPU Desktop* will appear. Click it to open the remote desktop in your browser.

Inside the desktop session, you can: * Open terminals * Run GUI applications * Use IDEs

The session behaves similarly to a Linux desktop computer running on the cluster.

You can transfer and manage files using: * The *Files* menu in Open OnDemand * SCP/SFTP

Your home directory on the cluster is available in the desktop session.

When finished:

• Save your work.
• Close applications.
• Return to the Open OnDemand dashboard.
• Select *My Interactive Sessions*.
• Click *Delete* to end the job.

Closing the browser tab alone may not stop the job.

The cluster is likely busy and is already using the GPUs for other running sessions. You can see what jobs are currently running by accessing the *Jobs* menu in Open OnDemand and selecting 'All Jobs'.

Possible causes:

• Browser popup blocker
• Expired session
• VNC/websocket startup failure

Try:

• Refreshing the page
• Relaunching the job
• Using a different browser

Interactive jobs automatically terminate when:

• Wall time expires
• The job is canceled
• Cluster maintenance occurs

Save work frequently.

* Request only the resources you need. * Use shorter wall times when possible. * Terminate unused sessions promptly. * Use interactive desktops for visualization and development, not large batch production runs.