Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
openmp on seaborg -- too good to be true?
0 views
Skip to first unread message
Jeffrey Hammel
Mar 13, 2004, 4:37:09 AM3/13/04
to
When I ran the openmp fish on seaborg, I get twice the speed of the MPI
fish and the pthreads fish with the number of openmp threads equal to the
number of pthreads equal to the number of processors for mpi. Anyone else
get this? Am I running this in a strange way to make the speed look
better, or is this due to a good implementation of OpenMP on the IBM SP?
fish and the pthreads fish with the number of openmp threads equal to the
number of pthreads equal to the number of processors for mpi. Anyone else
get this? Am I running this in a strange way to make the speed look
better, or is this due to a good implementation of OpenMP on the IBM SP?
Jeff
gennari
Mar 13, 2004, 5:28:32 AM3/13/04
to
We got similar runtimes for MPI and OpenMP until we hit a large number of
processors, at which point OpenMP failed. Did you use the same optimized
interact_fish code in both the MPI and OpenMP code? Did you turn compiler
optimizations on in both cases?
processors, at which point OpenMP failed. Did you use the same optimized
interact_fish code in both the MPI and OpenMP code? Did you turn compiler
optimizations on in both cases?
Frank
"Jeffrey Hammel" <jha...@landau.EECS.Berkeley.EDU> wrote in message
news:Pine.LNX.4.44.040313...@landau.EECS.Berkeley.EDU...
Jeffrey Hammel
Mar 13, 2004, 4:37:40 PM3/13/04
to
I used the default configure script and build method for homework 2.
Unmodified code. For some reason, OpenMP seems much faster.
Unmodified code. For some reason, OpenMP seems much faster.
Jeff
0 new messages