Author: Robert Hyatt
Date: 15:04:53 07/29/04
Go up one level in this thread
On July 28, 2004 at 12:15:21, Tord Romstad wrote: >On July 28, 2004 at 10:17:37, Anthony Cozzie wrote: > >>On July 28, 2004 at 04:44:51, Tord Romstad wrote: >> >>>On July 27, 2004 at 13:52:49, Anthony Cozzie wrote: >>> >>>>Perhaps that is a bit extreme :) There really aren't that many parallel >>>>programs out there: >>>> >>>>The Baron (dual only), Crafty, Diep, Fritz, Junior, Sjeng, Shredder, SOS, Zappa >>>>(very little testing) >>> >>>Amy, Hydra, Nejmet. I am sure there are others, but I can't remember any at >>>the moment. >> >>Hmm, I didn't know Amy and Nejmet were parallel. Do you know what method they >>use? > >No, I don't, but I'm fairly sure they are both parallel. > >>>I don't doubt that it is very difficult (although "one of the most difficult >>>programming tasks out there" is a pretty extreme claim) >> >>Well, first we must define "programming task" :) I do not include algorithms in >>this description. > >OK, that makes the claim a bit less extreme. :-) > >>In other words, I mean: given an algorithm, create machine >>code (through compilers if necessary) that implements it. In my opinion, >>parallel algorithms are many times more difficult than serial algorithms, > >Of course. > >>and alpha beta search is one of the more difficult ones. > >I find this a bit hard to believe, but I suppose I am not the right person >to speak with authority until I have given it a try myself. My impression >of computer chess in general so far is that everything is relatively easy >to implement (otherwise I wouldn't even have managed to write my program >in C), but that finding good ideas is very difficult. I find most of the >computer algebra stuff I do to be more difficult to implement than computer >chess algorithms by at least an order of maginitude. It is possible that >trying to implement a parallel alpha beta search will change my mind about >this, but so far I doubt it. I suspect it will. Alpha/Beta is, by nature, a serial algorithm. Parallelizing it is a royal pain... Parallelizing it _well_ is something beyond that even... > >The bottom line is probably that we are both too young and inexperienced >to judge which programming tasks are the most difficult. Let's meet and >discuss it again in 30 years or so. :-) I have 36 years of experience telling me how difficult it is compared to other parallel algorithms. :) > >>, but I think the >>>small number of parallel programs has to do with lack of interest as well as >>>difficulty. How many programmers have tried to parallelize their engines? >>>I doubt that the list would be much longer than the list of successful >>>implementations above. >> >>I think that is mainly due to a lack of access to dual machines. Most of the >>people who tried have been the ones with reasonably strong programs already, > >Yes, otherwise it would obviously be a waste of time. Unless you already >have a very strong program, there are easier and better ways to improve than >adding parallel search. > >>and no one who failed would be talking about it ;) > >I am not so sure about this. It seems reasonable that some of those who >try to write a parallel program would ask lots of questions here in the >process. If there were a lot of failed attempts to implement a parallel >search out there, we should have seen many such questions from people >who never managed to finish their task. > >Tord
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.