[CP2K:201] doing properties calculations in parallel...
hut... at pci.uzh.ch
Sat Jul 21 12:52:58 UTC 2007
> hi all,
> this was mentioned in a reply to some thread before, but i thought it
> might be better to discuss this independently:
> i was wondering, whether there would be a way to have some
> of the more time-consuming analysis, that is run during a quickstep
> MD, being done on a separate set of nodes. e.g. with the 'beloved'
> calculation of wannier centers. since that type of calculation is
> independent, one could just transfer the wavefunction, coordinates
> and then have a seperate communicator compute the centers, why the
> regular MD can just go on.
Although in principle a good idea, we have the problem that this
is rather difficult to implement. I think for this specific
case we should try to get better parallelisation for the individual
parts of the calculation.
> perhaps, there are other potential targets for that approach as
More information about the CP2K-user