Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Dr. Hsu @ Microsoft

Author: Robert Hyatt

Date: 15:43:15 10/17/99

Go up one level in this thread


On October 17, 1999 at 15:22:19, walter irvin wrote:

>On October 16, 1999 at 15:21:11, Bruce Moreland wrote:
>
>>On October 16, 1999 at 11:29:11, walter irvin wrote:
>>
>>>first i'm going to admit that i know as much about multi processing as i do
>>>about brain surgery .but it just seems to me that if a person had say 4 cpu's a
>>>way to attack the problem COULD be as follows .
>>>1.first divide the moves if there are 20 then each gets 5 ect
>>>2.after each cpu gets to say 5 ply take the high score and go with it .disconect
>>>the other 3 cpu's transfer them up the tree to 6 ply do the same thing over and
>>>over to get real deep ??????
>>>
>>>maybe if you could get deep enough , you could turn tactics into strategy just
>>>by sheer depth ???????
>>
>>It is more complicated than this.  The moves don't take the same amount of time
>>to search, some of them take a long time.  And this isn't just a hypothetical
>>case, it will happen in every search.  In a perfect world idle processors would
>>go help out on searches that are taking a long time.
>>
>>bruce
>
>ok how about instead take the start position give it to cpu A .now most programs
>crunch through 5 ply in a split sec .once cpu A  gets to 5 ply then  cpu B makes
>ply 4 its start position . when cpu B gets to 5 ply , cpu C makes cpu B 4th ply
>its start position . so on and so on . if it proves to be bad some where along
>the line simply change the direction of the search .in simple positions make the
>ply count higher , in complex positions make the ply count lower .it takes much
>less time to search from 1 to 5 ply than from 10 to 15 ply or 15 to 20 ply  i
>think .
>     there may be a major flaw in my suggestion????????? but i dont know ,maybe
>some one who does know could explain the flaw ???????????


there is a major flaw... it is called 'alpha/beta'.  and you are spending a lot
of time with one searching, and the other 479 doing nothing (DB2 used 480 chess
processors).



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.