GPU Node Access and Configuration Guide

This guide outlines the steps to access a GPU node with 8 NVIDIA A30 GPUs and configure a personal environment for Jupyter Lab. It’s intended for graduate students working with Dr. Femiani or other projects involving the MU Robotics program that require GPU resources.

Note: The node is currently reserved for specific use. In the future, it may be added to the RedHawk HPC Job Scheduler. This guide might not be updated at that time.

1. Establish SSH Connection with Port Forwarding

Set up an SSH connection with port forwarding to enable access to Jupyter Lab:

ssh -L 8888:localhost:8888 <userid>@mualhpcp50.hpc.miamioh.edu

• 8888:localhost:8888: This forwards port 8888 on the remote server to your local machine.
• Replace <userid> with your actual username.

Tip: Consider setting up SSH keys for passwordless login to enhance security and convenience.

2. Load Necessary Modules

Before running processes, load the required modules:

module load tmux-3.3a
module load anaconda-python3.10

To make sure these modules load automatically for future sessions, append them to your .bashrc file:

echo "module load tmux-3.3a" >> ~/.bashrc
echo "module load anaconda-python3.10" >> ~/.bashrc

Tip: Use source ~/.bashrc to reload the configuration without logging out.

3. Using Tmux for Persistent Sessions

Tmux allows you to run commands that continue even if you disconnect from SSH.

Start a new tmux session:

tmux

To reattach to a disconnected session:

tmux attach

Creating a New Tmux Tab for Jupyter

1. In tmux, create a new tab by pressing Ctrl + B, then c.
2. Name the tab by pressing Ctrl + B, followed by , and entering the name (e.g., “jupyter”).

4. Launch Jupyter Lab

Within the tmux session, launch Jupyter Lab without opening a browser:

jupyter lab --no-browser

Tip: You can also specify a different port (e.g., --port=8889) if needed to avoid conflicts.

5. Access Jupyter Lab

After launching Jupyter, control-click the provided URL to access Jupyter Lab locally, typically at:

localhost:8888

If you’re forwarded to a different port (e.g., 8889), follow the steps in Section 6.

6. Resolve Port Conflicts

If port 8888 is occupied and Jupyter defaults to a new port (e.g., 8889), set up a new SSH session with the updated port forwarding:

ssh -L 8889:localhost:8889 <userid>@mualhpcp50.hpc.miamioh.edu

Use this link to open Jupyter in your browser, or use tmux attach to find the session.

If the Jupyter link has scrolled out of view, create a new tmux tab and type:

jupyter notebook list

7. Set Up a Personal Conda Environment

You may want to use your own environment instead of the base conda environment. To create and clone an environment:

conda create --name myenv --clone base

Then, install your environment as a kernel for Jupyter Lab:

python -m ipykernel install --user --name myenv

8. Reconnect to SSH and Tmux Session

If disconnected, re-establish the SSH connection with port forwarding:

ssh -L 8888:localhost:8888 <userid>@mualhpcp50.hpc.miamioh.edu

Reattach to the tmux session:

tmux attach

9. Use the Web-based NX Client (Optional)

For interactive tasks such as downloading large datasets (e.g., from Google Drive), connect via the NX Web Client for a graphical desktop interface.

Instructions for setting up NX are available in the project Slack channel.