Author: Uri Blass
Date: 04:24:36 08/16/02
Go up one level in this thread
On August 16, 2002 at 07:00:40, Chuck wrote: >I can no longer find the exact text, but as I recall the author of Hiarcs 8 >claimed the search algorithm was greatly improved to avoid the "explosion" of >the search tree at great depths. This appears to me to be a false claim. > >I run alot of deep analysis with Junior 7, Fritz 7, Chess Tiger 14, Gambit Tiger >2, Shredder 6, Rebel Century 4, Gandalf 5 and Hiarcs 8. Often I let a program >analyze a single position for 2-4 days. All seem to handle this well except >Hiarcs, which is completely incapable. It seems there is a certain point where >the nps dramatically drops on Hiarcs. All the computers I use are 1.2 - 1.5 GHz >with 256-512 MB RAM, and Hiarcs will usually report 150-200 kn/s, but after a >day or more this figure will drop dramatically. Right now I'm looking at another >computer where it has been analyzing a position for 45:19:56 and it is reporting >23 kn/s! I've seen it drop further, to 10-15 kn/s where I kill it. This may >always happen (haven't really tracked it, normally I just give up on Hiarcs >after it wastes so much of my time). I used to and would like to still be a >Hiarcs fan, but this is a significant problem IMO. I do not have Hiarcs8 but I think that it is possible that it is a display problem(I know that a lot of chessbase engines have display problems and the only exception that I know is Deep Fritz). The problem is that the engine always assume that the number of nodes is smaller than 2^32 so the number of nodes is reduced to 0 after being 2^32-1 and if Hiarcs divide the number of nodes by the number of seconds it can get a very small number. The main question is if the branching factor of Hiarcs is increasing. I know that it was a problem with Hiarcs7.32. > >It seems to me the tree "explosion" is still a big, big problem and the author's >claim was false. My evidence suggests Hiarcs is crippled at deep analysis. I know that the problem for Hiarcs7.32 was a lot before getting 2^32 nodes. It is possible that the author did not expect people to use Hiarcs for many hours(and you need to use it for many hours to get more than 2^32 nodes). The problem of being crippled at deep analysis was a problem even at tournament time control on fast hardware and if the problem today is only at deeper analysis then it is wrong to say that the author's claim was false but only that the claim is misleading for analysis of many hours on fast pc. Uri
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.