Russell -
PMEMD 9 would require about 500 MB/cpu on 4 cpu's for a 700,000 atom
problem. I have run systems approaching 1,000,000 atoms; the current
limitation is a field size in the inpcrd file I believe. For pmemd at
least, assuming you have 1 GB/cpu, you can pretty much run whatever you want
up to the 1 million atom limit on 4 processors (and you want more processors
anyway for performance; the runtime is prettymuch proportional to system
size). The master node does require more memory, but the amount more is
typically not a huge deal. PMEMD prints out some estimates of memory usage
in the mdout, but these estimates don't include static allocations, code, or
stack allocations (ie., it is just the stuff dynamically allocated at
startup that is reported). The memory estimate I am reporting here is from
looking at a system process monitor under linux. Sander definitely requires
more memory than pmemd.
Regards - Bob Duke
----- Original Message -----
From: "Russell Brown" <Russ.Brown.Sun.COM>
To: <amber.scripps.edu>
Sent: Monday, May 07, 2007 6:36 PM
Subject: AMBER: PME memory requirement
> Can someone tell me how to estimate the per node memory requirement for
> the MPI version of PME? The molecular model for which I need to run
> simulations comprises 700,000 atoms. So for N MPI processes, how many
> bytes of memory will be used per process? And is there a master process
> for which the memory requirement is higher than for other processes?
> Thanks.
>
> --
>
> ------------------------------------------------------------
> Russell A. Brown | Sun Microsystems, Inc.
> russ.brown.sun.com | UMPK14-260
> (650) 786-3011 (office) | 14 Network Circle
> (650) 786-3453 (fax) | Menlo Park, CA 94025
> ------------------------------------------------------------
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.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 May 09 2007 - 06:07:22 PDT
