Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Cray Blitz

Author: Robert Hyatt

Date: 07:16:25 08/02/01

Go up one level in this thread


On August 02, 2001 at 09:31:34, Vincent Diepeveen wrote:

>On August 02, 2001 at 08:53:38, Robert Hyatt wrote:
>
>>On August 02, 2001 at 03:56:58, Dann Corbit wrote:
>>
>>>On August 02, 2001 at 03:32:12, Joshua Lee wrote:
>>>
>>>>Since Deep Blue isn't around playing games these days the only "Real"
>>>>Competition That is on the level of DeepThought or Blue would be Cray Blitz.
>>>>How does CB compare to the Deep's? I know DT could be searching 10Ply find a
>>>>move but a regular Fritz might take 20+ ply I am guessing that CB is closer to
>>>>the Deep's than your average program ...  What kind of differences between Cray
>>>>and Commercials are there in eval "IF"  that's a BIG "IF" eval is the main
>>>>difference between the two. How much memory went to Hash Tables etc? Is the
>>>>Depth = to Crafty only just deeper in searches? In other words is 12ply for CB
>>>>the same as Crafty? I played a game Between Fritz and Crafty at 12ply each and
>>>>then 12ply for Crafty and 14 for Fritz , Fritz lost both  not that this says
>>>>much as Fritz at 12 ply took tops 1:30 while Crafty took up to 30 minutes but at
>>>>14 Ply for Fritz it was not enough in the endgame. I don't know how much Depth
>>>>is good for fritz but it looks like 12Ply is a decent minimum for Crafty for
>>>>most positions except in the endgame.
>>>>
>>>>Also i asked about the games of Crafty Vs Cray Blitz when will these be
>>>>available?
>>>
>>>Cray blitz had memory bandwidth as a big advantage.  But personally, I think the
>>>8 way SMP boxes running Junior or Fritz would give cray blitz all it could
>>>handle.
>>>
>>>On the other hand, put crafty on a 64 way SMP Compaq Alpha machine, and only
>>>Deep Blue would have a chance against it.
>>>
>>>Pure speculation, of course.
>
>Is there any possibility to play a testgame(s) DIEP vs cray blitz
>somehow?

It is not easy to arrange.  I played 10 games vs crafty last year and crafty
did poorly.  No reason to believe Diep will do better.  And that was against
a "stock" cray blitz.  It would take 10 minutes to change it to use R=2 and
recursive null-move.  Another 10 to make it use the current adaptive R=3~2
null-move.



>
>If i understood well, you got like 11 to 12 ply with Cray Blitz,
>just what i get with diep at a dual machine?
>

11-12 plies with null move, R=1, non-recursive (only one null move allowed in
any single path between the root and the tip).



>>I would disagree.  Cray Blitz was very similar to crafty, other than it didn't
>>use recursive null-move or R=2.  But those would be tiny changes to enable.  And
>>it would then be searching just like Crafty, just at 7M nodes per second
>>rather than 1M...
>>
>>And CB had some eval things that I can't do in crafty due to memory bandwidth
>>limits...
>
>I do not see the connection between evaluation code and needing bandwidth
>for it. As far as i know evaluation has only to do with the way the
>processor is working. Isn't that the case here?


Nope.  Bandwidth to match patterns.  To load the entire contents of diagonals,
ranks, files.  To compute "quality mobility".  Etc.



>
>Best regards,
>Vincent



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.