Hi Albert,
A single calculation can use up to 16 cores of a node, although the latest development code uses threaded building blocks to let it scale to as many cores as are available in a node (for example, I am working now to get it running quickly on a Xeon Phi).
Sire does have MPI support, but this doesn't efficiently speed up waterswap so isn't used by the waterswap program. With multiple nodes, it is more efficient to run multiple copies of the simulation so that you get a good idea of the statistical error.
Best wishes,
Christopher
--
Sent from Gmail Mobile