Using the GPU cluster of CentraleSupelec

Licence : Creative Commons Attribution 4.0 International (CC BY-NC-SA 4.0)
Copyright : Jeremy Fix, CentraleSupelec
Last revision : January 23, 2024 09:11
Link to source : cluster.md
Lectures project page: https://github.com/jeremyfix/deeplearning-lectures/

To connect to the GPUs at CentraleSupelec, I strongly recommend you to use dcejs, a cross platform desktop application which allows to easily book a node and start a graphical session on the allocated node as illustrated below

Connecting to the GPUs during a lab session

During a lab session, you are provided with a reservation code that you must use for your labwork. The video below shows how to perform an allocation and start a graphical session with an example reservation called demoresa.

Connecting to the GPUs during free times to finish a labwork

You have the possibility to allocate a GPU for working after the official lab work sessions. In that case, you must allocate a node on the partition gpu_tp. Every session cannot last more than 2 hours. If you want to work for more than 2 hours in a row, you will have to make a new allocation once the first one ends.

Below, I show you how to allocate a node on gpu_tp for 2 hours.