Titan cluster in LPTMS is a Linux cluster composed of one head node and twenty-two compute nodes, it uses the scheduler “SLURM” to to keep track of available resources, allowing job requests to be efficiently assigned to various compute resources (CPU and GPU). The cluster provides computational capacity for single and multi-processor jobs, it possess 2914 CPUs, 3 GPUs and 8 TB memory, the highest CPU count per node is 192 and the maximum memory size per node provided is 3TB.

The head node is the server where you log in, compile code, assign tasks and coordinate jobs; the computing node is to perform the tasks through its memory and CPU.

A partition represents a group of nodes in the cluster. There are several partitions:

q-1heure     allows a maximum run time of 1 hour
q-1jour*     allows a maximum run time of 1 day
q-1sem       allows a maximum run time of 1 week
q-2sem       allows a maximum run time of 2 weeks
q-1mois      allows a maximum run time of 1 month
q-2mois      allows a maximum run time of 2 month
qbigmem      allows a maximum run time of 1 week, used for large memory work 
qopnmp       allows a maximum run time of 1 week, used for parallel jobs
debug        for the purpose of test

Using GPUs

Titan cluster contains five GPUs (Tesla T4 and NVIDIA L4), the GPUs are located in a different compute nodes. In these nodes, the Nividia CUDA Toolkit is installed and provides a compiler for NVIDIA GPUs, math libraries, and tools for debugging and optimizing the performance of applications.
Useful links:
Introduction to CUDA C++
PyCUDA
Numba

To demand the GPU resource for Python in the script:

!/bin/bash
#SBATCH --ntasks-per-node=1
#SBATCH --partition=gpu
#SBATCH --mem-per-cpu=10G
#SBATCH --time=1:00:00
#SBATCH --job-name=celloracle
#SBATCH --output=celloracle-%J.log

srun conda run -n tf python ./load-tensor.py

Using programming languages

Different languages are installed in Titan Cluster, like Python, C/C++, Fortran, Julia, etc.

Using SLURM

User Guide

Here is the detail about Titan Cluster.