I have a heterogenous HPC cluster; half the nodes have 56 cores, the other half has 48. When a job runs on a 56 core node, the results vary slightly from when the same job runs on a 48. This slight difference has a major impact on business decisions however.
I have tried using job templates to specify the number of cores a job should run it with, but the job fails if the numcores figure does not match the number of cores the node has.
So my question is, is there a way in HPC to make the node present a specified number of cores, less than what it actually has, to run jobs with it, preferably without doing anything in the node BIOS or some hardware fix to achieve this?
Or any other way to make this work, so maybe some combination of job template config plus some HPC setting so jobs only use the number of cores specified in numcores, no matter how many cores a node actually has.