Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Faster Hardware benefits slow searchers or fast searchers more???

Author: Dave Gomboc

Date: 08:02:18 10/01/99

Go up one level in this thread


On October 01, 1999 at 05:26:54, Peter McKenzie wrote:

>On October 01, 1999 at 02:47:06, Dave Gomboc wrote:
>
>>On September 30, 1999 at 20:26:02, Peter McKenzie wrote:
>>
>>>On September 30, 1999 at 14:35:25, Charles Unruh wrote:
>>>
>>>>
>>>>  In the past i thought faster hardware would benefit slow searchers like
>>>>Chessmaster more than fast searchers like Fritz.  Now i'm more inclined to
>>>>believe that it makes more sense that faster hardware benefits fast searchers
>>>>more.  For the reason that positional ideas are for the most parts moves made
>>>>from practical experience/knowledge, that we can't always quite calculate.
>>>>However, faster hardware gives programs the ability in many instances to
>>>>actually be able to calculate the result.  So although i think Chessmaster is a
>>>>truly awesome engine especially against programs running up to 233Mhz I expct
>>>>that on a P450Mhz it will come in 3rd or 4th.
>>>
>>>This isn't a simple question.
>>>My basic take on the issue is that programs that sacrifice some speed (in terms
>>>of NPS) in return for a better evaluation will do better as hardware speed
>>>increases and/or time controls get longer.  The basic premise behind this
>>>argument is that an extra ply of search depth becomes less important at greater
>>>depths, so at greater depths the evaluation function becomes a more important
>>>factor.
>>
>>And the basic problem with this take is that the premise, at least by today's
>>best guess, simply wrong.
>>
>>ref: Crafty Goes Deep, DarkThought Goes Deep (both in ICCA Journal)
>
>I'm not convinced by the methodology used in the 'Go Deep' papers.  But
>certainly the premise could be completely wrong :-)
>
>One interesting point, as discussed in the DartThought articles, is that most
>published self play experiments have been statistically insignificant.
>
>Another thing, perhaps fixed ply games aren't what we should be interested in
>since in real games searches are much deeper in the endgame.  Fixed time would
>be better.
>
>>
>>Dave

I agree that fixed time would be better.

BTW, I'd cringe if I saw results posted based on Auto-232 games! ;-)

Dave



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.