Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crap statement refuted about parallel speedup

Author: Vincent Diepeveen

Date: 18:49:52 09/24/01

Go up one level in this thread


On September 23, 2001 at 23:09:08, Robert Hyatt wrote:

>On September 23, 2001 at 19:38:11, Vincent Diepeveen wrote:
>
>>On September 23, 2001 at 18:47:27, Dave Gomboc wrote:
>>
>>>On September 23, 2001 at 18:32:50, Vincent Diepeveen wrote:
>>>
>>>>The average test done here at CCC is 10 to 20 seconds a move.
>>>>
>>>>Even the new try to make a WACII the dudes test at most at 1 minute
>>>>a move.
>>>>
>>>>So you're kind of wrong here.
>>>>
>>>>Most tests in ICCA and advances in ICCA are based upon anything under or
>>>>equal to 8 ply.
>>>>
>>>>I can only remember a single article 'crafty goes deep' where some
>>>>deeper searches were done.
>>>
>>>That may the speed they are measuring at, but the assumption is that the speedup
>>>rate will remain roughly constant as the depth increases further.
>>
>>well that assumption is wrong. Read what bob writes about Cray Blitz
>>in journal in ICCA.
>>
>>Also note that speedup of cray blitz is reported to be 3.9 there at
>>4 processors, doing that by moving 64KB blocks or whatever and nearly
>>a fullwidth search with singular extensions!
>>
>>At 2 processors cray blitz gets *easily* 1.9 speedup.
>
>
>
>Don't mix apples and oranges.  the 1.9 was without SE.  That speedup was
>done at the time of my dissertation.  Not with the SE version of CB.
>
>
>>
>>With singular extensions i do not even get *close* to that speedup. I
>>get 1.8 then.
>>
>>Why wouldn't i manage at just 2 processors to get 2.0 then at 2 processors
>>if i turn off singular extensions at 3 minutes a move?
>
>Simple answer. You would have to have _perfect_ move ordering to do that.  I
>don't believe it is possible.  Except in very rare positions.  If your move
>ordering is not perfect at a split node, that kills any hope of a perfect 2.0
>speedup.

So you say kind of here that it is NEVER possible in your viewpoint to
get a 2.0 speedup when only measuring search times?

Please give a clear statement here.

Yes or No.


>
>
>
>>
>>You guys are not listening to arguments like: "game space".
>>
>>What the hell do you know from computerchess anyway?
>>
>>I mean you *never* of course measured anything. So arguments
>>i write down here and reasons simply do not get listened to!
>
>So long as you refer to super-linear speedups nobody is going to listen.
>It simply isn't possible as the sequential algorithm can be speeded up
>first...
>
>This is old news.  It is well-known.  Anybody worth their salt in computer
>science is going to remind you of it over and over and over...
>
>Eventually it will "sink in".
>
>
>
>
>>
>>>Dave



This page took 0.1 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.