Author: blass uri
Date: 22:52:17 05/28/98
Go up one level in this thread
On May 28, 1998 at 18:39:41, Christophe Theron wrote: >Sorry Georg, but unless I understand nothing about computer chess >programming, when the hash table is full the program DOES keep on >analyzing. Maybe you get a few percent slowdown, but the program in no >way stops analyzing! > >I guess the manual just warns you that you should have more memory to >get ABSOLUTE optimum performances at longer time controls, but with less >memory you are just a few percent under this optimal curve. > >That's all. > >This issue has been discussed several times here. Maybe Fritz manual >warns you too much with too heavy words, but thinking that the analysis >nearly stops is a common mistake among computer chess users. I wonder >why... > >Anyway, you can be sure that your 32Mb is not that bad. > >With my program, Chess Tiger, I usually get a few percent speedup a long >time controls when I increase my hash table size from 16Mb to 32Mb. What >a big deal! > >BTW, I have read several times here that Fritz takes advantage of more >hash tables, and need them in fact, but I don't remember somebody kind >enough to post here some real data. Isn't time to show some numbers so >this legend can be verified? I would have done it myself, but I don't >have Fritz5... in the past I posted here some real data that proved that sometimes it is better that fritz5 will have small hash tables. from my experience when I increase the hash tables fritz5 usually analyze more selective lines for the same brute force depth. the result is that sometimes fritz5 needs more time to analyze at the same brute force depth with big hash tables and if to discover something the brute force is the important thing fritz5 may be faster with small hash tables. In the example I posted fritz5 was faster with smaller hash tables because of another reason(with small hash tables it analyzed the mistake it wanted to play at depth 15 for a long time and saw this is a losing move when with big hash tables it chose not to analyze the move at depth 15 for a long time and discovered only at depth 16 it is a losing move, but usually this is not the case. the position I posted: r4rk1/2p3pp/p7/1p1pq3/8/2P2N2/PPQ2KPP/R1B5 white to move. the move white played in the game(correspondence chess) was 19.Kg1 and the result was that he lost after 19... Rxf3 20.gxf3 Qe1+ 21.Kg2 Re8 22.Kh3 Re5 23.Bf4(the best move) Rh5+ 24.white resigned I tested fritz5 with different hash tables to see the time it takes it not to play 19.Kg1 and the results were that with 10000Kbytes (it cannot have exactly 10000Kbytes) it was faster than with 10 times bigger hash tables to find 19.Qd3 I have a computer with memory of 256 Mbytes so I tried the maximal hash tables that I can give fritz5 and it was even slower. it needed 90 minutes in order not to play 19.Kg1 (I do not remember the exact time but it was near 90 minutes) in this case it did not see at depth 15 19.Kg1 lose but saw at depth 15 that the move 19.Be3 is better than 19.Kg1 it considered 19.Be3 for a very long time at depth 15 and this is the reason it took fritz5 90 minutes. with 100000 Kbytes it finds 19.Qd2 after something near 60 minutes at depth 16 while with 10000 Kbytes it finds 19.Qd2 at depth 15 after less than half an hour. I have pentium 200MMX. Fritz5 manual warns me too much. They say: "Please note that it does not make a lot of sense to make the program run for many hours to analyze a position if the hash tables are full. The extra time hardly compensates for the dramatic reduction in speed. It may be more adventagous to use the "Correspondence analysis" function..." I do not agree the correspondence analysis can miss Rxf3 and suggest 19.Kg1 so it make a lot of sense to make the program run for many hours to analyze a position. I know that I should not believe to what chessbase say about fritz5. fritz5 is not bad at longer time control my nunn match(1000 seconds per move) proves this fact and I do not understand why chessbase make the impression that if I want to use the computer for hours it is better to use another program. another interesting thing is that when fritz5 finds Qd3 it evaluates the position as adventage for white when black can force draw after 19.Qd3 by 19...Rxf3+ and if 20.Kxf3 Rf8+ 21.Kg4 h5+ 22.Kh4 when 22...Rf5 or 22...Rf1 leading to draw. > >If somebody posts real numbers, I will do the same kind of experiment >with Chess Tiger and others programs I have. > > > Christophe
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.