Author: Robert Hyatt
Date: 11:08:45 01/29/02
Go up one level in this thread
On January 28, 2002 at 16:57:35, Joshua Lee wrote: >>>Congrats on taking the initiative. >>> >>>Can we assume that this is a DOS program and that it is related somehow to the >>>Russian mainframe chess program of the old days? >> >> >>Even less than you can assume Crafty is related to Cray Blitz. :) > >This is basically the version that was rewritten into "Turbo-C" and from the >documentation has refinements from the version that played in the 2nd Computer >Olympiad in London 1990. It is about as close as we can get to the actual >mainframe version from the 70's. So like Crafty is related to Cray Blitz , so is >this PC version of Kaissa. I would like to mention something fairly obvious but >yet interesting that is where Kaissa in 1972 played Komsomolskia Pravda Readers >it took 90 minutes in one instance to search 1,500,000 this PC version on a >1Ghz it searched 2,673,745 in 56 seconds. Huge increase from 200-300 Positions >per second i'd say. The readers played Spassky in 1971 in two games and drew one >and lost the other. Spassky at that time was 2690 which would put the readers >average at 2490. Kaissa managed the same result against the readers losing one >game and drawing the other which would put it's average at 2290. The problem is that the way they searched in 1974 has _nothing_ to do with the way they searched in 1990. Ditto for Blitz in 1977 WCCC, vs Cray Blitz in the 1983 WCCC event. There is simply nothing comparable between those two programs, even though I wrote _both_. Faster hardware completely changed the way the search was used... And it changed what could be evaluated as well... > >Since Kaissa was using upwards of 3 hours on some moves just to search 3million >positions if this PC version does the same but with so much less time you could >say that by estimation this version is atleast that elo strength at time control >on a 1Ghz, also if given the same amount of time if the gain in elo would remain >constant at 90 per tripple of clock speed it would play at 2650 where current >programs do this with alot less time even on this hardware. >For example Fritz 6 managed an average of 2551 on a P3 500. Fritz in 1972 would >have still been better if my estimates are correct but hey this is just my >opinion. I doubt it. In 1972 a brute-force program had _no_ chance against a good selective search program. Because of the very slow hardware. Try a depth of 3-4 plies for Fritz and you will see what I mean. It will play like a complete idiot. In 1972 a _selective_ program was getting 4-5 plies _max_. Brute force only became workable with the faster computers from CDC/Cray that came along in 1975 or so...
This page took 0.09 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.