From d3e102 %-% at %-% ames.pnl.gov  Thu Apr 22 04:21:20 1993
Date: Thu, 22 Apr 93 11:21:20 -0700
From: d3e102 "-at-" ames.pnl.gov
Subject: Benchmark Info. Available
To: chemistry[ AT ]ccl.net
Message-Id: <9304221821.AA07611&$at$&ames.pnl.gov>



Recently, in regard to the availability of Gaussian on high-end PC's, someone
mentioned that it would be helpful to have comparative information available on
the relative performance on other machines.  Here in the MSRC we have just
completed a benchmark study of Gaussian 90 and 92, as well as many other
popular ab initio packages (GAMESS-US, GAMESS-UK, HONDO, DISCO, MOLPRO
and ACES II).  We examined their respective performance in terms of CPU time
and wall clock time on a variety of machines, including workstations from Sun,
HP, IBM and SGI, as well as large machines like a Cray Y-MP and a C90.  

In an attempt to avoid skewing the results towards one particular type of
calculation, we included 5 different molecules (ranging from small, with high
symmetry to large with little or no symmetry).  Among the methods examined
were RHF (conventional and direct), UHF, MP2 (conventional and direct), MP4,
gradients and hessians for RHF and MP2, CISD, CCSD, QCISD and CASSCF.

A small amount of information was included on the recently introduced DEC Alpha
workstations.  Lastly, there is a reference to several density functional
programs which we have also benchmarked, using some of the same molecules and
similar quality Gaussian basis sets.

This information is contained in a Pacific Northwest Labs technical report
which is slowly winding its way through clearance.  We will make it available
to anyone requesting it and a PostScript version will be available for 
downloading via anonymous FTP.  The latter has not yet been set up.
 
People will have to judge the cost effectiveness of workstations vs super-
computers vs PCs for themselves.  However, I can't help but note that the
quantity which is of primary interest to most scientists using these programs
is the "time-to-solution".  Very fast, very expensive computers are nearly
always oversubscribed.  This results in lengthy waits in long queues for a
chance at the limited resources.  Our experience has been that it is often
possible for the unemcumbered tortoise to beat the overburdened hare.

David Feller
Molecular Science Research Center
d3e102 -A_T- pnlg.pnl.gov