There are three timeshare machines available for GPU processing: blaze, cream, and vanilla.
blaze: 4x Titan X cards, 64GB RAM
cream: 4x Quadro RTX 6000 cards, 375GB RAM
vanilla: 4x Quadro RTX 6000 cards, 375GB RAM
Access
To access any of the timeshare machines, you first need a Computer Science Account. If you do not have one, apply for one here.
You won’t be able to ssh directly to any of the timeshare machines from outside of Computer Science, you will first have to log in to a gateway machine:
knuckles.cs.ucl.ac.uk – Taught students
tails.cs.ucl.ac.uk – All other students and staff
To do so, ssh from your local machine using your Computer Science username (not your UCL username), for example:
ssh jbloggs@knuckles.cs.ucl.ac.uk
From knuckles/tails, you can then ssh on to one of the timeshare machines:
ssh jbloggs@cream.cs.ucl.ac.uk
Running Code
It is important that you only use one GPU per user. To do so, you must source the script detailed in the log-on message:
source /usr/local/cuda/CUDA_VISIBILITY.csh
Python
The default Python in your environment when you log in will be 3.6.8, you can check this by running the following:
python3 --version
There are scripts to modify this in /opt/Python, named Python-3.x.x_Setup.csh. To modify your environment, use the source command to run the corresponding setup script:
source /opt/Python/Python-3.11.5_Setup.csh
You can check this has worked by running python3 --version again.
For any issues regarding storage or software please contact us.