Author: Christophe Theron
Date: 11:40:36 06/03/98
Go up one level in this thread
On June 01, 1998 at 11:45:24, Thoralf Karlsson wrote:
>On June 01, 1998 at 06:09:29, Moritz Berger wrote:
>
>
>>a) we don't know if the other programs got less than 64 MB RAM (this
>>might have been different for individual matches, for all we know, e.g.
>>Nimzo getting 200 MB hash tables vs. Fritz with 44 MB HT - but the point
>>is: we just don't know)
>
>
>Some facts about RAM-size used by SSDF until 98-02-22:
>
>64 MB was first used during the period May -97 and beginning of Sept
>-97. Hiarcs 6 was the first program to get this possible advantage.
>Until the latest list from February 22, 64 MB was partially used also by
>Rebel 9, MChess Pro 6.0, Genius 5, Nimzo 98 and Shredder 2.
>
>The use of RAM 98-02-22 can be seen in the table below:
>
>
> 96MB 64MB 48MB 32MB
>
>Rebel 8 0% 0% 0% 100%
>MChess Pro 7.1 0% 0% 3% 97%
>Genius 5 0% 0,5% 0% 99,5%
>MChess Pro 6.0 0% 7% 0% 93%
>Hiarcs 6 0% 12% 5% 83%
>Rebel 9 0% 10% 8% 82%
>Nimzo 98 0% 55% 7% 38%
>Shredder 2 0% 33% 30% 37%
>Fritz 5 18% 82% 0% 0%
>
>With 48 MB of RAM Shredder 2 can, as far as I know, use the same amount
>of RAM for it's hash- and evaluation tables as what is recommended for
>64 MB, so 48 MB seems to be as good as 64 MB.
>
>The two programs Nimzo 98 and Shredder 2, which we tested just before
>Fritz 5,
>had more than 32 MB RAM in more than 60% of the games. Since a couple of
>months ago all testers have 64 MB RAM. (The tester with 96 MB has
>stopped testing, so the average RAM for Fritz 5 will decrease) Because
>of that the percentage of 64 MB will rise. All new programs will be
>tested solely with 64 MB RAM.
>
>When we were going to start the automatic testing of Fritz 5 late
>January we were in a process of upgrading the RAM-size, and as is shown
>above several programs had already used 64 MB of RAM in rather high
>percentages. We believed that Fritz 5 behaved equally as the other
>programs when it received more RAM, that is, we expected a speedup of
>6-7% for a doubling of the RAM-size, which possibly could give 4-5
>points extra. In this situation we thought it was acceptable to use
>Fritz 5's autoplayer, which worked only with 44 MB of hash or more.
You were completely right. I believe all sensible people should
understand that the hash table advantage results in a very small
strength advantage, and which is probably ridiculous when compared to
the possible rating error.
So you have our complete support for what you do. I speak for myself, of
course, but I'm sure lots of reasonnable people around will agree.
Christophe
>If I have understood the recent discussions correctly, it doesn't seem
>that Fritz 5 earns so much more than the other programs, when the
>hash-size is doubled.
>
>In possible future editions of the SSDF rating list, the average amount
>of RAM for programs tested on P200 MMX will be shown.
>
>Thoralf Karlsson
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.