Pete is right, but note that pmemd scales better than sander and newer
versions of
sander scale better than older ones. NAMD is great for very large
systems when one
wants to run a very standard MD simulation. The philosophy of sander is
to support a wider
variety of calculations, perhaps with some scaling cost but that is
being improved all the time.
for a recent quantitative comparison of amber, charmm and namd see this
link:
http://amber.scripps.edu/amber8.bench2.html
note that both the scaling and wallclock run time for amber8 at SDSC is
BETTER for amber
than namd 2.5 on blue horizon, but namd is better on data star. This
also shows how much
amber has improved in scaling and single processor speed as new versions
have come out.
the message is that all of these codes are constantly improving and you
really need to check current
data and test the programs on your own systems. relative code
performance depends on the
machine architecture, program versions and the system that you are
simulating.
carlos
Peter Gannett wrote:
>We've note similar results though our turn around comes after 8 CPU. My
>understanding is that in addition to what Carlos says, amber does scale
>directly with # of CPUs. Note to be hieratic but, NAMD seems to scale
>better and can run with the amber force field.
>
>Pete
>
>cripps.edu
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Jan 12 2005 - 15:53:00 PST
