Skip to end of metadata
Go to start of metadata

Introduction

This document attempts to provide clarity on different task options such as  --ntasks versus  --ntasks-per-node  and --nodes .

Documentation

The use of --ntasks will place tasks on available nodes for throughput purposes.

If you request --ntasks=16 and there are 3 nodes with 8 CPUs and a total of 16 available free CPUs then your job will be allocated to those free CPUs.

Additionally, if there are any minimum node requirements then --ntasks will understand and accommodate such a requirement. Meaning it will allocate the minimum number of nodes necessary to satisfy the scheduler.

The use of --ntasks-per-node will place the same number of tasks on each node requested.

Additionally, if there are any minimum node requirements then --ntasks-per-node will understand and accommodate such a requirement. Meaning it will allocate the minimum number of nodes necessary to satisfy the scheduler.

The use of --nodes by itself has a default of --ntasks-per-node=1.