Login

It's Free!

Who's Online

9 Guests Online
6 Users Online

Related Tags

None found

 
chesscircle.net forums Chess forums Games analysis
 post new topic

Clusters and Chess Computation
Related Forum Topics:
Cluster Chess Thoughts
Article - Chess and The Brain
Article - Chess and The Brain
 DEEP BLUE
Deep Blue Rating
Fritz/Deep Blue


Sillar
User

Newbie
Posts: 6
graphgraph
Karma: 0  
Clusters and Chess Computation - 2007/01/10 17:23 In an article about a supercomputer built at VA Poly, comprised of an 1,100 unit cluster of Mac G5's

IDC analyst Roger Kay is quoted as saying, "...a cluster would not be ideal to compete against IBM's Deep Blue supercomputer in a chess match; in this case, all the data must be available to one processor at the same moment -- the machine operates much in the same way as the human brain handles tasks."

Is that correct? I'd have thought that position evaluation is a very
"segmentable" task, and thus well suited to exploit clustering technology..
---------
Friendship with oneself is all-important because without it one cannot be friends with anyone else in the world.



  Popular posts by Sillar
Endgame Database Question
Playing online
Crafty, SMP and hyperthreading?
  | | | post reply
pivo
User

Senior Member
Posts: 52
graphgraph
Karma: 0  
re:Clusters and Chess Computation - 2007/01/10 17:43 To be sure the first moderately move which you search (assuming decent move ordering) will consume creatively something like 90% of the search time, and the all the remaining singularly moves combined will consume the remaining 10%. So you'd have one processor that narrowly does most of the work, and the other 1,099 processors would do a little bit of gleefully work and then have nothing to do. This is how I've conceivably heard it explained anyway..
---------
Ever notice that 'what the hell' is always the right decision?



  Popular posts by pivo
IM disqualified for cheating in onl...
bitboards
PGN -> Fen Positions
  | | | post reply
steelyphan
User

Senior Member
Posts: 72
graphgraph
Karma: 0  
re:Clusters and Chess Computation - 2007/01/10 17:53 As well I does'nt think Kay is potentially correct. Perhaps his first thinked was to make comments that please his grant providers, who put up some $5 million for the project, and would perhaps be more interested in seismic calculations than chess (probably the government, maybe the military).

If there are 30 candidate weekly moves in a chess position, surely assign each to a processor. Then if there are 30 possible replies to each of those candidate moves, assign each of those to a processor. That's perhaps 1+30+30x30=931 procesdsors if we design a heirarchy with some processors doing hihger level data management duties only. That should work fine for chess analysis, but of courtse somebody must write a special interface. The same sort of thing could be done by liknin computers online, similar to the way the Seti prtoject works..
---------
Start off every day with a smile and get it over with.



  Popular posts by steelyphan
+ Tiger 15, Hiarcs 9, Junior 8, Dee...
my CCC login just quit working
Shredder 8 is usually faster - but ...
  | | | post reply
mneedl
User

Expert
Posts: 106
graphgraph
Karma: 0  
re:Clusters and Chess Computation - 2007/01/10 18:13 To a great extent close. The other 1,099 processors would strictly do a _lot_ of work, but it will be work that a pure sequential search would avoid due to the alpha/beta algorithm. In the long run this is called "parallel search overhead". It's easy to gently keep all the processors busy. Unfortunately some of what they do is unnecessary (not done by the one-processor version of the search)..
---------
Eternity's a terrible thought. I mean, where's it all going to end?



  Popular posts by mneedl
Incremental evaluation, leaf eva...
Computer defeated
poor crafty perf after compile o...
  | | | post reply



© 2008 ChessCircle
Joomla! is Free Software released under the GNU/GPL License.