Author: Vincent Diepeveen
Date: 15:30:42 10/04/03
Go up one level in this thread
On October 04, 2003 at 17:53:07, Russell Reagan wrote: You are comparing apples and beans here. If you do a cns search you have a lot of overhead which depth limited alfabeta programs do not have. If you use additional computer power to make up for that, then that is very trivial to me. the importance is the research itself, not how many processors were needed for it to look good. It searched just above 1 million nodes a second against diep in all the games played. That was up to 200 processors. Not 3 - 3.5 mln. Perhaps it did that in tactical testsets though... Comparing with top programs is not so interesting. Interesting is how it would have done with diep's evaluation function. Basically p.conners is the best show that good tuning and adjustment and testing of a program is more important than a very deep search is, as it didn't search deep. >On October 04, 2003 at 14:54:39, Omid David Tabibi wrote: > >>Going back to McAllester and Yuret's 1993 article, Alpha-Beta Conspiracy Search, >>the idea seems very interesting. Apparently this idea hasn't received much >>attention, and the only top engine which uses a conspiracy search is P.ConNerS. >> >>Has anyone tried the ABC algorithm? Any experimental results eversince the >>original publication in 1993? > >http://www.uni-paderborn.de/fachbereich/AG/monien/PERSONAL/FLULO/Lehre/ParProg/CUMPI/pconners.html > >"P.ConNerS runs in parallel and gets its playing strength out of a >workstationcluster, which consists of 160 Pentium II, 450 Mhz Processors. Those >are connected with a new european interconnection network, the so called SCI >network. On that machine P.ConNerS examines between 3.5 and 5.0 Mnds/sec. On the >160 processors it achives a speedup of about 50." > >Surely this means that the 50x speedup results in the program searching 3.5 - >5.0 Mnps, and not that the program searches 50 x 3.5 or 5.0 Mnps. > >It is interesting though. I wonder how this program, running on fast hardware, >would compare with other top programs on the same hardware. From the page above, >it gives a good result against GMs, but it seems like other programs had much >better results (P.ConNerS lost to two GMs, while several other top programs have >finished GM tournaments undefeated). Based on this alone (which is pretty >shaky), I'd guess it's probably around the strength of a lot of the strong >amateur programs. Of course, most of those strong amateur programs aren't >running at 5Mnps either, so it may be weaker. Pure speculation of course :)
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.