Hi Bob,
Just as a follow up, even though the login and compute nodes had the same specs, it seems that the login node limited ram per user to 5GB and failed while the compute nodes I would use the specified 24gb and it worked perfectly. So looks like a memory issue after all.