[Beowulf] VASP

Anthony J. Ciani tony at ciani.phy.uic.edu
Thu Aug 5 14:37:42 PDT 2004


On Wed Jul 14 15:50:36 PDT 2004 Mike Davis wrote:

> Does anyone on the list have any experience with VASP on Athlon? The 
> number from the VASP site look good, but I'm looking for a Computational 
> Chemist with some real life experience.

Plane wave codes such as VASP run very well on Athlon machines.  The
Athlon/Opteron chips have three floating point units compared with the
Xeon's two, so a 2.0GHz Athlon performs like a 3.0GHz Xeon.  However, the
Athlons tend to have lower cache sizes and smaller memory bandwidths than
Xeons, which degrades their lead.  Opterons, on the other hand, have the
same basic internal structure as the Athlons, but they have much better
memory band width and branch prediction.  A 2.0GHz Opteron beats a 3.4GHz
Xeon.

The best F90 compiler to use for Athlons is Intel's fortran compiler
(ifort), and if you're non-commercial it's free too.  Even though it's
32-bit ifort works great for Opterons too.  The best for Opterons is made
by PathScale, but it costs $$$$, and it's performance isn't much better
than ifort.  PGI is OK for both, but be weary of the "extras" as they tend
to be buggy or slow.  There are others, but not really worth mentioning.

> Also does anyone know of a running version on VASP on Mac OS X?

Yes, I have this working on a dual 2.0GHz G5.  The performance is also
comparable to a 3.4GHz Xeon.  I recommend using XLF to compile (it ain't
free $$$).  I sent the makefile to the VASP maintainer, or you could
modify your own.  The thing to watch out for is the case-insensitive file
system.  Of course you could reinstall as (or maybe migrate 
to) case-sensitive.

-- 
------------------------------------------------------------
              Anthony Ciani (aciani1 at uic.edu)
           Computational Condensed Matter Physics
   Department of Physics, University of Illinois, Chicago
              http://ciani.phy.uic.edu/~tony
------------------------------------------------------------




More information about the Beowulf mailing list