Author: Robert Hyatt
Date: 08:07:30 04/22/02
Go up one level in this thread
On April 21, 2002 at 02:43:41, Joshua Lee wrote: >In a rather lengthy post about how Deep Thought Dominated the ACM events you >came up with an 86% score for DT against it's opponents. This was deep thought vs Micros, I believe, and included events from roughly 1987 thru 1995. >I am curious as to if >we could compare PC's to this what kind of system would we have to score 86% >against? Pentium 90? Wchess and Fritz were on P90's and I am not sure about the >various other programs like from the 1994 ACM. Obviously it is pretty impossible >to say well you need this kind of system to run this commercial software to >equal DT, because nobody seems to agree how much knowledge and what knowledge >was contained within that Program; We can't say well Crafty or Fritz has to >search exactly the same ammount of nodes because they might have to search alot >less. So to rephrase if DT was 86% better than it's competition(315elo) to have >a program that strong compared to DT does it have to have a system only as fast >as that Deep Thought system or can it be any Setup that perform's 315 points >higher than DT's competition on a Pentium 90? I don't really see how to do this. We have the following data: 1. Deep Thought produced a 2650 rating vs human GM players in 40 move/ 2hr games (25 consecutive games). 2. Deep thought dominated computers in general. How you connect 1 and 2 seems difficult. I, for example, believe that comp vs comp games are far different animals than comp vs GM games. IE it is possible based on the above that the micros were playing at a 2400 level back then, based on 86% being about 250 Elo (very roughly). However, it could be that the micros would have been worse than that against humans, or (not likely) that they might have been even better than 2400. I don't see how to deduce any information about 2 above, from 1. To make up the 250 rating point shortfall predicted by the 86% victory margin suggests that the PC programs needed hardware at least 32x faster than the best they had back then. That would be five doublings to gain about 250 Elo. IE assuming a P6/200 in 1995 (I believe that is about when the P6/200 came out, I could be off a year however so that maybe we should assume a P5/133 in 1995) we would need a machine running at somewhere betweeh 133*32=4.3ghz up to 200*32=6.4ghz, depending on which processor speed you take. We are roughly there for multiple-cpu machines, not quite there for singles. Of course that is also for deep thought speed, where DB2 was well over 100 times faster than the fastest deep thought ever used. > >Maybe that is more ethics , is it fair kind of question. for example if their >system was equivalent to a Dual 2Ghz but we need a Quad 4Ghz system to get the >same performance. I don't think anyone has ever thought about that. > >Also how would we compare DT, it's 86% score was against various types of >hardware, possibly some that are faster than current Desktops. I don't think so. I believe I only counted micro programs. Which did include some hot micros like the PA family from HP and also microprocessors from Sun, etc. But I believe I factored out things like the Cray, HiTech (itself a hardware solution) and a few things like the CM-5. > >Lastly if we compare their data from a recent pdf file about DB, They said DBjr. >lost 2 out of 30 games 315elo higher than a P200 (maybe P90) that would put >their single chip at 2795 on the SSDF list depending on which top program from >that hardware you compare it to. > >Thankyou Remember also that their "single chip" machine was running at 1/10th its normal speed, as mentioned by Hsu and Campbell at several of their talks...
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.