SLURM

Initiate and manage SLURM tasks

Contents
Most used parameters Example by executing a simple script GPU memory selection options

Most used parameters:

Parameters	Description
#SBATCH --ntasks-per-node=2	# Number of tasks per phisical CPU core
#SBATCH --time=1:00:00	# Script duration (days-hrs:min:sec)
#SBATCH --job-name=test_job	# Job name
#SBATCH --mem=1G	# Ram memory for rendering (e.g. 1G, 2G, 4G)
#SBATCH --error=testerror_%j.error	# Print the errors that occur when executing the job
#SBATCH --cpus-per-task=1	# Number of processors required for a single task
#SBATCH --output=testoutput_%j.out	# Print the results from scripts and the values it returns
#SBATCH --gres=gpu:2	# Number of cards per one nod allocated for the job
#SBATCH --nodelist=cuda4	# Executing on specific nodes, e.g. cuda4 is for executing only on cuda4 host

export PATH="/opt/anaconda3/bin:$PATH"

source /opt/anaconda3/etc/profile.d/conda.sh

conda create -n virtualenv python=3.8

conda activate virtualenv

echo "FINKI FCC"

Example by executing a simple script

#!/bin/bash

#SBATCH --job-name=test_job

#SBATCH --time=1:00:00

#SBATCH –-ntasks-per-node=1

#SBATCH --error=testerror_%j.error

#SBATCH --output=testoutput_%j.out

export PATH="/opt/anaconda3/bin:$PATH"

source /opt/anaconda3/etc/profile.d/conda.sh

conda create -n virtualenv python=3.8

conda activate virtualenv

echo "FINKI FCC"

The script is executed via sbatch <scriptname>.sh

GPU memory selection options

There are 4 options for selecting GPU memory and this can be done by combining some of the commands in the script

GPU Memory	Code for the script
16 GB GDDR6	#SBATCH --gres=gpu:1 #SBATCH --nodelist=cuda1 (or cuda2 or cuda3)
32 GB GDDR6	#SBATCH --gres=gpu:2 #SBATCH --nodelist=cuda1 (or cuda2 or cuda3)
48 GB GDDR6	#SBATCH --gres=gpu:1 #SBATCH --nodelist=cuda4
96 GB GDDR6	#SBATCH --gres=gpu:2 #SBATCH --nodelist=cuda4