[gmx-users] How many processors do I need?...

[Mark Abraham]

Brian Wengerter wrote:
> Hi all,
> 
> Sorry for asking such a naive question, but I didn't get any good 
> results searching the lists.  I'm interested in modeling a system that 
> would contain a lipid bilayer and, at most, two proteins of moderate 
> size.  I'm looking to build a cluster to perform the simulations, but I 
> don't have a good idea as to what would be a reasonable number of 
> processors for a system of that approximate size.  Could someone suggest 
> to me what would be appropriate?  Basically, I'd like to know what I 
> need to get, and I'd also like to know how much it'll cost.  :-)  
> Thankfully, the latter is less of a concern, but I'd rather not waste 
> money, either.

Unless you plan on maintaining a dedicated computing cluster in the 
medium term, your best option is likely to be getting access to time on 
some regional or national computing facility's hardware. That allows the 
people who know about hardware and software to have reasonable machines 
and to install the programs properly and frees you up to concentrate on 
the science. The cost for that probably varies widely, but I can't 
imagine a DIY job by someone of low experience getting better value for 
money than such facilities. For example, in Australia, the 
government-funded APAC National Facility (http://nf.apac.edu.au) grants 
time freely to academic projects on a merit basis, in lieu of giving 
academic grant money to individual research groups to purchase computing 
resources. To get access all you need is someone at a university and a 
reasonable project :-)

The simulation you describe sounds pretty large - while (for example) a 
small-to-medium cluster of modern Linux boxes running GROMACS, using MPI 
  and commodity networking would probably get the job done in some 
reasonable fashion, the central facility has far better scope to scale 
up should the problem demand it.

Mark