Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hiarcs 7.32's Depth compared to Deep Thought

Author: Robert Hyatt

Date: 18:31:38 08/24/00

Go up one level in this thread


On August 24, 2000 at 15:41:17, Joshua Lee wrote:

>On August 24, 2000 at 10:14:05, Robert Hyatt wrote:
>
>>On August 23, 2000 at 23:20:35, Joshua Lee wrote:
>>
>>
>>Where did I say that?
>
>Say what??
>
> And how could we answer that, since Deep Thought's
>>hardware was faster than _any_ PC-type machine available today, even including
>>machines like the 8-way xeons.  And that is only deep thought.
>
>How is their hardware faster but their search speed slower?? Was there that much
>more knowledge in DT's software than in readily available software now?
>>
>>I simply said that I would take _their_ 12 ply search over _any_ 12 ply search
>>by another program.  They didn't use null-move or any other speculative form
>>of forward pruning.  The only mistakes they made were horizon mistakes that
>>everybody makes, and they made fewer than the rest due to their search
>>extensions.
>
>I have an idea what about crafty? You know your software better than anybody
>else, How is it different from Cray Blitz and Cray Blitz to DT if that makes it
>easier?


Crafty and Cray Blitz are very similar.  Different approaches (Crafty uses
bitmaps, CB did not except in rare places).  Similar evaluations and searches
except Crafty uses recursive null move, CB didn't.  CB used singular extensions,
crafty doesn't (yet).  Both did the usual check and recapture extensions, both
used fractional ply extensions, etc.

DT was faster than CB.  It didn't rely on null-move, but was otherwise a pretty
similar sort of program, excepting evaluation.  DT was faster than any micro
today, when you count NPS with chess engines.  They did no selective tricks,
which means when they searched 12 plies, they saw _everything_ within 12
plies.  When they extended to 20 plies, they saw everything along their
variations.  No moves casually discarded here and there...

That is why I would take their 12 ply search over any other program's 12 ply
search today.  I don't know of any program that searched _full width_ to 12
plies today, with no pruning tricks of any kind (including null-move, etc.)

Given the choice, which would you take, a 12 ply search that relies on tricks
to get there, or a 12 ply search that relies on fast hardware to get there?
DT pretty well proved that the latter was the right way to go, for them.



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.