AMBER Archive (2007)

Subject: Re: AMBER: Hardware recommendations please

From: David LeBard (david.lebard_at_asu.edu)
Date: Thu Jun 07 2007 - 13:31:21 CDT

Hi Jeremy,

You must take into consideration the size of the system you would like to
simulate before asking if you can run AMBER on a single PC. For instance, I
am able to run AMBER rather well on the newest dual core PC on a box of 500
water molecules, and in several days I can generate nanoseconds of data.
However, if I wanted to simulate something really big and interesting, like
a membrane bound protein or even just a small biomolecule in a typical
amount of explicit solvent molecules, a single PC will not cut it.

For our group, we study molecular liquids and small biomolecular systems,
but we still had to build a linux cluster for some simulations, and use the
bigger supercomputing centers when our system size grew larger than our
small cluster (38 procs). Indeed, you can buy cheap nodes for roughly $1-2k
nowadays, but if you want a quality cluster you will invest most of your
money in your switch and interconnects. Of course it is possible to build a
cluster capable of running medium sized simulations only using gigibit
ethernet backbone, but the parallel scaling isn't wonderful (as previously
discussed) and you may be waiting a long time for your data. That said,
there are places were you can buy cheap pre-built "clusters" for about $15k,
where there are 5 little nodes, each with 4 cores, and a huge amount of
total hard disk space and memory. Still, they generally run over gigE and
it might be a pain if one of their little parts fails and you have to go
digging in there with your miniature tool set. But I digress...

So, to answer your question properly: It is (in my opinion) ridiculous to
consider simulating biomolecules on a single PC, but very small system with
about 1000 or so atoms will not be a problem at all.

Best of luck,

David LeBard
Graduate Student
Arizona State University

On 6/7/07, Jeremy Weedon <Jeremy.Weedon_at_downstate.edu> wrote:
>
>
> Thank you. Please pardon the ignorant question, but is it ridiculous to
> consider trying to run Amber on a single PC?
>
> JW
>
>
> Knut Langsetmo <knut_at_bbri.org> wrote on 06/06/2007 01:16:10 PM:
>
> > hello jeremy,
> >
> > for the best bang for the buck, you might check out:
> > http://www.symmetriccomputing.com/
> >
> > with shared memory and infiniband interconnects you can get good
> > performance scaling with additional cpus (as opposed to the problems
> > with scaling on multiple cpus with ethernet connections, currently being
> > discussed on the list).
> >
> > On Jun 6, 2007, at 12:15 PM, Jeremy Weedon wrote:
> >
> >
> > Hi folks,
> >
> > We're brand new to Amber, some hardware advice would be much
> > appreciated. We plan to install to a desktop PC, probably running
> > some variety of Linux. We need to figure out what make/model, how
> > much RAM, how much diskspace, whether we need RAID, multiple CPUs
> > etc. Are some Linux distros more suitable than others? Should we
> > invest in a high-end graphics board?
> >
> > TIA,
> >
> > Jeremy Weedon, Ph.D.
> > Senior Statistician
> > Scientific Computing Center
> > State University of New York
> > Downstate Medical Center Box 7
> > Brooklyn NY 11213
> > +1 (718) 270-7424
> >
> > --knut
> >
> > Boston Biomedical Research Institute... Today's Research for
> > Tomorrow's Health.
> > Please visit us at www.bbri.org

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu