5. Experimental Apps (Unsupported)

To start a Jupyter notebook with a GPU

Either: Visit the dashboard at:

• https://gpucluster1.doc.ic.ac.uk/pun/sys/dashboard/

or

• https://gpucluster4.doc.ic.ac.uk/pun/sys/dashboard/

After logging in:

• Choose Jupyter from 'Pinned Apps'

or

• Click 'My Interactive Sessions' in the main menu along the top, choose Jupyter from the left pane

Example screenshot:

Edit the options where required, a functioning python environment is required, so make sure you have run these commands or similar on a lab PC:

https://www.imperial.ac.uk/computing/people/csg/guides/python/virtual-environment/

mkdir /vol/bitbucket/$USER
export PENV=/vol/bitbucket/${USER}/myenv
python3 -m virtualenv $PENV

## Install jupyter notebook dependencies
pip install jupyter jupyter-lab notebook

Jupyter Interface

• choose Notebook or Lab

Environment Setup for Custom Python

• enter the correct path to the created environment, eg:

/vol/bitbucket/${USER}/myenv

If the path is not correct or valid, the default read-only env /vol/bitbucket/starter is loaded

Preferred Shell

• Leave as default or change to a different value, make sure the shell executable is valid, eg.

  /bin/bash or /bin/csh or /bin/zsh
  

Email

To receive email notifications of job start and completion, tick the box next to:

'I would like to receive an email when the session starts'

When you have chosen appropriate options:

• Press Launch

After a short wait, you will have a notebook job in the queue that will last for 4 days.

Optionally, ssh directly using a Terminal or use VSCode to connect, the hostname is listed in the job details as above

eg.

ssh username@hostname.doc.ic.ac.uk