FSL -
Simply, I am running a temp concat group melodic with 272 inputs. I've had a couple failures on 32GB servers. As such we've moved to our super-computing institute to a machine with 1100 cores and 3TB of globally shared memory.
As far as I can tell it seems that the group melodic is a single threaded process. Is there a way to parallelize the PCA portion of the process and utilize multiple cores? Will giving melodic 600gb of memory allow it to run faster, or will it simply prevent it from running out of memory during the PCA?
|