[Beowulf] Re: Barcelona vs. Woodcrest, computational chem

Guilherme Menegon Arantes garantes at iq.usp.br
Wed Sep 26 10:21:56 PDT 2007


On Wed, Sep 26, 2007 at 09:03:10AM -0700, beowulf-request at beowulf.org wrote:
> 
> andrew holway wrote:
> > Multi reference quantum chem. Gaussian and Molcas.


Hi Andrew,

You still need to tell us more about the jobs. I run large active space
with large basis set CASPT2 jobs in MOLCAS. Since it's coded rather
inefficiently, and the matrixes are huge, it doesn't not fit in memory
and scratch files take up to 500Gb. So, you really need a 64bit machine
with the fastest disk system you can get. It will be disk/IO bound and
the processor will matter little.

Also MOLCAS is a serial code in any Multi-ref part, so no use for extra 
cores/processors.
Gaussian is rather limited for Multi-ref QM. So, I guess you will either
use it for other kinds of calculations or will gave up using it in favor
of only MOLCAS (or MOLPRO, GAMESS, etc...).


> As usual, real use case tests are the most realistic predictor of 
> performance.


Joe is right here, but sometimes the hardware is just not avalaible to
testing.

Regards,

G
--

Guilherme Menegon Arantes, PhD       São Paulo, Brasil
______________________________________________________




More information about the Beowulf mailing list