Author: Mike S.
Date: 06:31:27 09/02/01
Go up one level in this thread
On September 02, 2001 at 07:11:57, Uri Blass wrote: >It seems to me that people simply look at the number of >nodes per second to decide which program is more knowledge based >program and it is clearly wrong. >(...) >There are cases when the fast searchers have >more realistic evaluation and it is not clear to me if less cases >then cases when the slow searchers have more realistic evaluation. Maybe it's not even correct to call the low nps programs slow searchers. They reach similar depths as the high nps programs do, with less nodes . I think they are considered "knowledge"-programs because obviously they do heavy pruning of some kind which means they probably use much knowledge for that. But this hasn't to be traditional chess knowledge... I remember Christophe Théron said, that there is something like computer specific chess knowledge IIRC, used for algorithms. But much of that this is certainly something we won't find in a chess book. It's astonishing how different the node rates are, comparing programs of similar strength. The bandwidth is approx. 1:8 (maybe even 1:10). Some time ago, I have done the following comparison on P3/700, 32 MB Hash each (Grundstellung = starting position): ---------- P3/700 MHz The King 3.12c (CM 32 MB Hash): 0:56 4431056 (Grundstellung) = 79,1 kN/s 1:17 5818502 (nach 1.a4) = 75,6 kN/s Hiarcs 7.32 / 32 MB Hash: 0:29 2247 kN (Grundstellung) = 77,5 kN/s 0:36 2741 kN (nach 1.a4) = 76,1 kN/s Fritz 5.32 / 32 MB Hash: 1:03 38877 kN (Grundstellung) = 617,1 kN/s 0:25 15663 kN (nach 1.a4) = 626,5 kN/s Junior 5.0 / 32 MB Hash: 1:36 50302 kN (Grundstellung) = 524,0 kN/s 0:47 25265 kN (nach 1.a4) = 537,6 kN/s ---------- As an additional information, these figures show that Fritz 5.32 used around 1100 Pentium CPU cycles per node, and Hiarcs 7.32 or King 3.12c use 9100 cycles per node. Regards, Mike Scheidl
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.