My understanding is that the --ntasks-per-node directive applies to each job 
rather than to the array as a whole. To make each job in the array require 
exclusive access to a compute node, I think you want to replace 
--ntasks-per-node=1 with --nodes=1.


I found Harvard University's documentation on job arrays helpful.

https://rc.fas.harvard.edu/resources/running-jobs/#Job_arrays


> sbatch --array=1-10  --ntasks-per-node=1 -o env.log-%A.%a
>  --wrap="/usr/bin/env"
>
> I'd like to distribute each task on a different node but it looks like
> it is using fill up the node before moving on to the next node.

Reply via email to