Hi Ilyas,
On an AMD Opteron system the architecture specific compilers from the Portland Group (PGI) will probably do much better than the generic gfortran. You may want to benchmark that option. I use Opteron exclusively with PGI compilers and in my systems pmemd runs faster than sander.mpi. The only caveat is that I do not run this comparison across a network, but only within a multicore CPU - I don't have fast network interconnects so I avoid that. I use MPICH2. PGI compilers are not free, unfortunately. Since one can't compile amber -static (to load all libraries), I can't send you the compiled pmemd and sander.MPI from my PGI compiled AMBER 10 because it won't run on your system w/o the PGI libraries. However, the following should be possible to achieve a benchmark. Go to the PGI web site and download their software with a trial license. Then you can compile, benchmark and if things are good you can decide to buy.
Cheers, Sergio
Sergio Aragon
Professor of Chemistry
SFSU
-----Original Message-----
From: Ilyas Yildirim [mailto:i-yildirim.northwestern.edu]
Sent: Friday, March 25, 2011 1:54 PM
To: AMBER Mailing List
Subject: [AMBER] AMD Opteron system - compiling pmemd with intel or gfortran?
Dear All - I am trying to compile pmemd in an AMD opteron cluster. AMBER 9
and 10 and pmemd are compiled using gfortran. I benchmarked sander.MPI and
pmemd using exactly the same conditions (core #, local disk, etc). The
test jobs finished in 91 and 118 minutes, respectively, for sander.MPI and
pmemd. Now, this is a very surprising result because in all the intel
based clusters I have worked on, pmemd was almost 1.2-1.3 times faster
than sander.MPI.
I am not the admin of the cluster and do not have much flexibility on what
to install to the system. I was planning on compiling the intel compilers,
openmpi, and amber9 on my local directory to see if intel is going to do a
better job than gfortran for pmemd. The cluster is a little bit messily
organized and all the mpi/lib files are put into local places like
/usr/bin. Namely, I am having trouble using - for instance - intel
compiled openmpi on pmemd installation.
Anyways, my question is if it is worth trying all this hassle in and AMD
Opteron system or is pmemd really not efficient in this type of system. I
checked out the mailing list but could not find an answer to this question
(or maybe I missed them). There is quite some discussion between intel vs
gfortran on pmemd, but did not see anything connected with AMD Opteron
systems. Any idea/suggestion/comment is well appreciated. Thanks in
advance.
Best regards,
Ilyas Yildirim, Ph.D.
-----------------------------------------------------------
= Department of Chemistry - 2145 Sheridan Road =
= Northwestern University - Evanston, IL 60208 =
= Ryan Hall #4035 (Nano Building) - Ph.: (847)467-4986 =
=
http://www.pas.rochester.edu/~yildirim/ =
-----------------------------------------------------------
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Mar 25 2011 - 17:00:02 PDT
