[molpro-user] Cross-node parallelization
grhmrugg at gmail.com
Mon May 1 11:18:37 CEST 2017
Dear Molpro Users,
We have two questions,
Firstly, does molpro’s CCSD(T) routine make use of global arrays or
openmpi? We are using the 64 bit MPP version of Molpro.
Secondly, should this routine in theory be parallelizable across multiple
nodes using infiniband? We are trying to run large CCSD(T) jobs in parallel
across multiple nodes. We are only using ~1/2 of the cores on each node, as
the RAM demand is sufficiently high that using all cores on a given node
would overshoot the node’s available RAM.
In our test case, we attempted to allocate ten cores per node by specifying
them in the nodefile. However, we observed the calculation actively slowing
down when more nodes are made available. Is this to be expected, or have we
likely made an error in our system setup?
Many thanks in advance for your help,
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Molpro-user