Re: [sdpd] Multi-core and SDPD

[Favre-Nicolin Vincent <vincefn...@users.sourceforge.net>]

> Are some of you already using multi-core PCs for applications
> in crystallography specially adapted, and which ones ?

   Not me. But for direct-space SDPD, you do not necessarily need a multi-core 
application, as long as the algorithm is 'stupid' and does not learn from 
other structures - like most [reverse] Monte-Carlo methods. Running 20 
parallel, independent optimisations is the same as running a single one with 
a 20x processor. Which was why Fox cannot parallelize its calculations, from 
design choices - something I regret now, though it simplifies code and 
optimizations.
   Of course, it *would* be better to use a more intelligent algorithm that 
would learn from its errors ! It would definitely be a plus for genetic-like 
algorithms, etc...

   An interesting article about using clusters (among other things, that's in 
§3.2) is Shankland et al., Z. Kryst. 219(2004), 857
http://www.extenza-eps.com/extenza/contentviewing/viewJournalIssueTOC.do?issueId=2599

Afterthought : this one is more complete
http://scripts.iucr.org/cgi-bin/paper?hx5019

	Vincent
-- 
Vincent Favre-Nicolin
Université Joseph Fourier
http://v.favrenicolin.free.fr
ObjCryst & Fox : http://objcryst.sourceforge.net


 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/sdpd/

<*> To unsubscribe from this group, send an email to:
    sdpd-unsubscribe...@yahoogroups.com

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/