[Beowulf] updated GPU-HMMer/mpiHMMer bits
landman at scalableinformatics.com
Sun Feb 8 21:11:42 PST 2009
This time compared to a loaner AMD shanghai 2.3 GHz (same one I did the
rzf tests on a few weeks ago).
From the mpihmmer mailing list:
A new release of GPU-HMMER is available at www.mpihmmer.org. The most
notable change in the new code is support for multi-GPU systems. We
have tested the current GPU-HMMER with up to 3 GPUs, and have achieved
over 100x speedup with sufficiently large HMMs. A few bug fixes have
been applied as well, so I would encourage users to update. Users who
update should be aware that several command line options have changed,
and should check the GPU-HMMER user guide for details
While the system requirements haven't changed from the last version,
users who intend to use multiple GPUs should be aware that they will
need a substantial amount of system memory in order to do so. The 3 GPU
system I've been using has 16GB RAM. This is probably a bit overkill,
but 8GB or so would probably be appropriate.
As always, any comments, bug reports, etc. are welcome.
Updated mpiHMMer results were shown at SC08, the graph I saw showed
maxing out about 180x over a single thread, though I think the parallel
IO version can scale higher still.
Cudos to JP and the team for doing a great job on this!
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web : http://www.scalableinformatics.com
phone: +1 734 786 8423 x121
fax : +1 866 888 3112
cell : +1 734 612 4615
More information about the Beowulf