I've been trying figure out what is going on and part of the problem seems to be a buggy scheduler, which cannot be remedied at the moment. E.g. 64G maxvmem is reached in one subject within 12 seconds (when running a batch of over 100 participants). Running it again later (in a very small batch of 3 subjects) results in a succesful processing at maxvmem of 19G..

The scans we use have these dimensions, are they in any way problematic?