[gmx-users] memory requirements

[Syma Khalid]

Hi Erik,

Many thanks-it would be good to have the capabilities to simulate big
systems so will go for 1GB per core.

-Syma

***********************************************************************
Dr Syma Khalid 
RCUK Fellow (Chemical Biology)              
School of Chemistry
University of Southampton	
Highfield			 	
Southampton      phone: (0)2380-594176
SO17 1BJ          email: S.Khalid at soton.ac.uk    			
U.K.
***********************************************************************
 

-----Original Message-----
From: Erik Lindahl [mailto:lindahl at cbr.su.se] 
Sent: 20 February 2008 22:51
To: syma.khalid at bioch.ox.ac.uk; Discussion list for GROMACS users
Subject: Re: [gmx-users] memory requirements

Hi,

On Feb 20, 2008, at 10:42 PM, Syma Khalid wrote:
>
> I am about to buy some compute nodes for a mini-cluster (mainly for  
> running
> gromacs simulations).
> I have just been asked:
> 'One factor that needs to be considered is how much memory should be
> supplied per node, I suspect the choice is between 4GB & 8GB per  
> node (ie.
> 0.5 or 1 GB per processor-core for an 8 processor-core node).'
>
> by our HPC technicians. Any advice as to which I should go for, the  
> 4 or 8
> GB version?


It depends on the type of molecules you intend to simulate.

If it's going to be small systems (tens of thousands of atoms) you  
should be fine with 512MB per core, but for more general stuff I'd go  
with 1GB per core (that's what we recently ordered).

Since a bit of memory will be used by the OS, file system buffers, and  
executables I would guess the memory "available" to simulation data  
would be a factor ~3 higher in the latter case.

Cheers,

Erik
------------
Erik Lindahl   <lindahl at cbr.su.se>  Backup: <erik.lindahl at gmail.com>
Assistant Professor, Computational Structural Biology
Center for Biomembrane Research, Dept. Biochemistry & Biophysics
Stockholm University, SE-106 91 Stockholm, Sweden
Tel: +46(0)8164675  Mobile: +46(0)704218767  Fax: mail a PDF instead