[Beowulf] Barcelona vs. Woodcrest, computational chemistry research

[richard.walsh at comcast.net]

-------------- Original message -------------- 
From: Joe Landman <landman at scalableinformatics.com> 

> Hi Andrew: 
> 
> andrew holway wrote: 
> > Multi reference quantum chem. Gaussian and Molcas. 
> 
> Without doing a benchmark it will be quite hard to indicate which one 
> may be faster. 
> 
> Barcelona is a quad core with a good memory system (relative to the 
> Clovertown). If your problems are large enough to spill to disk for 
> matrix elements, then it isn't likely that the CPU performance would be 
> the dominating factor. If you have enough ram that your problem remains
Gaussian and GAMESS algs. typically do not store integrals unless they fit into memory.  It has been more efficient to just recompute them than to store them.  So large memory is very good, but ultimately there is never enough, so you often recompute.  Some routines parallelize well and can use dense matrix routines as I recall.  So if you were working on smaller systems at modest levels of theory you could put all the integrals in core and bandwidth would be very important. If you were working with larger systems and higher levels of theory you probably would have to resort to recomuting the integrals and therefore would be less concerned about bandwidth and more focused on clock and pure floating point.   Your probably job mix and benchmarking it will be key in making the right choice.
It looks like dual socket Barcelona bandwidth on stream will be around ~15 Gbytes/sec compared to about 9 Gbytes/sec for Harpertown.  You can consider this for a bandwidth intense workload (latencies are ~70 to ~110 nanos, respectively).  On the other hand, Harpertowns Linpack numbers are much better than Barcelona's if you are have to compare 2.0 GHz Barcelona to a 3.0 GHz Harpertown.  For two sockets the numbers at these clocks I have seen are 77 Gflops vs 45 Gflops (Harpertown to Barcelona).  As I suggested Barcelona may have clocks faster clocks by Q4 and this would change this a good bit.  These are just guidelines to consider in selecting systems and workloads to >>benchmark<<.  
Hope this is useful ...
rbw
-- 

"Making predictions is hard, especially about the future." 

Niels Bohr 

-- 

Richard Walsh 
Thrashing River Consulting-- 
5605 Alameda St. 
Shoreview, MN 55126 

Phone #: 612-382-4620
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20070926/bbaceac5/attachment.html>