Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The overall performance of Gandalf 4.32 f is 2545 too low !

Author: pavel

Date: 10:00:23 11/01/00

Go up one level in this thread


On November 01, 2000 at 12:36:43, Uri Blass wrote:

>On November 01, 2000 at 11:26:47, Peter Skinner wrote:
>
>>>I think on a AMD K6-2 450 MHz is the rating from the SSDF OK, but with the next
>>>games is the rating higher, I am sure. The AMD processor is not the best
>>>processor for Gandalf 4.32f !
>>>
>>>Best
>>>Frank
>>
>>Not only that, but in my testing, Gandalf 4.32e is much stronger than Gandalf
>>4.32f.
>>
>>In 5 min games. Total games 85:
>>
>>Gandalf 4.32e: +56 =12 -17
>>Gandalf 4.32f: =17 =12 -56
>>
>>In 15 min games. Total games 75:
>>
>>Gandalf 4.32e: +47 =13 -15
>>Gandalf 4.32f: +15 =13 -47
>>
>>In 40 min games. Total games 50:
>>
>>Gandalf 4.32e: +26 =15 -9
>>Gandalf 4.32f: +9 =15 -26
>>
>>In 40/40 games. Total games 60:
>>
>>Gandalf 4.32e: +43 =8 -9
>>Gandalf 4.32f: +9 =8 -43
>>
>>Each series was started with a new book. Both programs on a PIII 500, 256 megs
>>ram. 56 meg hash tables for both. Here is the performance of white/black via
>>elo_stat:
>>
>>White Perf.  : 49.4 %
>>Black Perf.  : 50.6 %
>>
>>Gandalf 4.32e was the strongest version, and should have been the commercially
>>released version. Here is a set of games in the Fritz interface:
>>
>>These were 40/40 games: Total 150.
>>
>>Gandalf 4.32e: +87 =23 -40
>>Fritz 6a: +40 =23 -87
>>
>>These were 40/40 games. Total games 150.
>>
>>Gandalf 4.32f: +53 =47 -50
>>Fritz 6a: +50 =47 -53
>>
>>You can see the "e" version destroyed Fritz 6a, and actually was the strongest
>>version over ICS play as well. My question is why wasn't it the commercial
>>version?
>
>Your results suggest that 4.32e is more than 100 elo better than 4.32f at all
>time controls and even more than 200 elo better than 4.32e at 40/40.
>
>They suggest that 4.32f is 125 elo better than Fritz6a.
>
>If you are right then the only reason that I can find not to sell 4.32e is if
>the programmer want to use it to win a correspondence championship but the first
>thing to suspect is that something is wrong with your results(maybe Fritz has a
>bug that tells it to play weaker only in eng-eng matches against Gandalf4.32e
>and maybe the machine that you used for 4.32f had problems and 4.32f was slowed
>by a significant factor)
>
>Uri

unless the results (or similar results) are confirmed by several others, who
have messed with 4.32e too, I wuold'nt even drop my pants ;)

Pavel



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.