Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: speedup of cray blitz as published in 1997

Author: Robert Hyatt

Date: 06:53:27 09/25/01

Go up one level in this thread


On September 25, 2001 at 07:21:29, Vincent Diepeveen wrote:

>On September 25, 2001 at 00:38:36, Robert Hyatt wrote:
>
>>On September 24, 2001 at 22:30:26, Vincent Diepeveen wrote:
>>
>>>Hello,
>>>
>>>Here written down speedups as claimed by a guy called R. Hyatt
>>>by cray blitz for 24 different positoins as they occured in
>>>a game:
>>>
>>>pos  speedup
>>>1     2.0
>>>2     2.0
>>>3     2.0
>>>4     2.0
>>>5     2.0
>>>6     2.0
>>>7     1.9
>>>8     2.0
>>>9     2.0
>>>10    2.0
>>>11    2.0
>>>12    1.9
>>>13    1.9
>>>14    2.0
>>>15    2.0
>>>16    1.9
>>>17    1.7
>>>18    1.8
>>>19    2.0
>>>20    2.0
>>>21    2.0
>>>22    1.9
>>>23    2.0
>>>24    2.0
>>>avg   2.0
>>
>>Please re-do your math.  No way to average 2.0 with even _one_ value that
>>is < 2.0
>>
>>There are several above that are less than 2.0
>
>Page 16 at ICCA journal March 1997 you claim
>Average for 2 processors 2.0

OK.  rounding.  the actual value above is 40 - 1 / 20 = 39/20 = 1.95.




>
>>>So YOU, Robert Hyatt, claims in an OFFICIAL magazine,
>>>called ICCA journal march 1997,
>>>an AVERAGE speedup of 2.0 with cray blitz at 2 processors.
>>>
>>
>>I don't see an average of 2.0...
>>
>>
>>
>>>Now i claim the same with DIEP if i'm not using dangerous
>>>extensions (which btw are turned on by default).
>>
>>No... You claim > 2.0 which is not possible over a set of positions.  Unless
>>your sequential search is simply badly flawed.  Then it doesn't matter.  Go get
>>a copy of my Ph.D. dissertation.  I proved quite clearly that for a minimax
>>search (alpha/beta with worst-possible ordering) it is possible to get a
>>near-perfect speedup curve.  I also proved that with perfect ordering, the
>>same is possible.  But The real point was that I clearly proved that we can't
>>get perfect move ordering (for obvious reasons) so that for normal cases, an
>>optimal (4.0 using 4 cpus) is not possible.
>
>The only time when i got close to 4.0 at 4 processors was when i
>forward pruned in a very dubious way. That was at world champs.

I get close to 4.0 all the time.  Just not _every_ time.  The average is in the
3.2-3.3 range for test sets, maybe 3.2-3.5 for real games.  But there are those
exceptions that drag the average down.


>
>Nowadays versions i'm pretty sure i'm not even close to the 3.7 which
>you had with Cray Blitz. Of course at the Quad i only ran with versions
>which have turned on dubious extensions by default.
>
>I have no idea what a run longer than say 10 minutes gives for kind of
>speedup, but i can imagine very well that because of reasons written down
>here, that it won't be good.
>
>For parallellism, running on 2 processors is perfect!


1 is even better.

there will _never_ be any "search overhead".

>
>>
>>
>>
>>
>>>
>>>It appears you hadn't turned them on either (smart guy
>>>to publish only speedups without dangerous extensions and only
>>>tell in 2001 that you hadn't turned them on).
>>
>>I was supposed to tell I hadn't turned on something I was not using because
>>it caused a serious problem???  I don't follow that logic.  And I don't consider
>>the SE as "a dangerous extension".  I don't do "dangerous extensions" anywhere
>>in my code, either now or then...



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.