AMBER Archive (2005)

Subject: RE: AMBER: meaningful tax for sander question

From: Ross Walker (ross_at_rosswalker.co.uk)
Date: Tue May 31 2005 - 14:55:14 CDT


Dear Luis,
 
There are benchmarks for amber available on the amber website - see
http://amber.scripps.edu/amber8.bench1.html
 
As for a test case the dhfr example should definately tax your system. If
you want to make the dhfr benchmark in $AMBERHOME/benchmarks/dhfr take
longer then you can just increase the number of steps that are run.
 
Edit bench.dhfr and change nstlim from say 100 to 1000, this will take
around 10 times longer.
 
If you really want to test your cpu and memory hard I would try the
factor_ix benchmark in the benchmarks directory. This has 90906 atoms which
is big. It should also take a lot longer to run by default. Again if you
want longer just make nstlim in the run script bigger.
 
Note, using PMEMD in place of sander for these tests will probably stress
your hardware more, especially in parallel.
 
I hope this helps
All the best
Ross
 

/\
\/
|\oss Walker

| Department of Molecular Biology TPC15 |
| The Scripps Research Institute |
| Tel:- +1 858 784 8889 | EMail:- ross_at_rosswalker.co.uk |
| http://www.rosswalker.co.uk/ | PGP Key available on request |

Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.

 

  _____

From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of
Cordova, Luis E.
Sent: 31 May 2005 12:20
To: amber_at_scripps.edu
Subject: AMBER: meaningful tax for sander question

Hi,

 

I am trying to benchmark/tax an experimental processor running Amber8, in
particular the sander link.

Does anyone know/could point me to other meaningful examples or benchmarks
that tax the sander link significantly?

Some example that is reported on the literature with results and can be
extended in size and also studied in regards to its scalability varying for
instance the number of processors. One example coming with the source code
of Amber is /benchmarks/dhfr but it takes only seconds to run on a single
processor and there is no documentation for how to make it bigger or how to
extend it or if it is meaningful to the biochemistry/&c. community.

The problem has to tax the processor preferably while going through the
Ewald and FFT parts of the sander link.

I am particularly interested in problems which demands resources or time
beyond realistic computational constraints.

 

Many thanks, and I will post back the responses I get,

 

\Luis

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