Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: crafty 16.3 - something broken in K and P endgames?

Author: Robert Hyatt

Date: 10:44:21 02/07/99

Go up one level in this thread


On February 07, 1999 at 12:30:06, James B. Shearer wrote:

>On February 05, 1999 at 17:58:27, Robert Hyatt wrote:
>
>>On February 05, 1999 at 00:13:59, James B. Shearer wrote:
>
><snip>
>
>>>              time=120:00  cpu=99%  mat=0  n=436256725  fh=4%  nps=60589
>>>              ext-> checks=49824007 recaps=1196664 pawns=18327198 1rep=14431267
>>>              predicted=0  nodes=436256725  evals=82091710
>>>              endgame tablebase-> probes done=0  successful=0  maxd=32
>>>              hashing-> trans/ref=32%  pawn=1%  used=w99% b99%
>>>White(41): quit
>>>execution complete.
>>
>>
>>this says something is wrong.  pawn=1% means you are running about 1/2 the
>>speed you normally should.  Check the size of hashp as it is either bogus or
>>something else is broken.
>>
>>fh=4% is impossible.  That means that in only 4% of the nodes where the search
>>fails high it fails high on the first move.  The lowest value I have ever seen
>>here is 88%, and it is normally above 90%.  Is this a stock crafty, or one you
>>compiled yourself?  IE it is possible that the compiler is breaking something.
>
>       I am using wcrafty_16.3.exe downloaded from your site.  I downloaded it
>again and the copies were identical.  I think hashp is ok, most positions pawn
>is circa 95%.  Is crafty supposed to be completely determinstic (except for
>slight timing differences) on a 1-cpu system?  I tried running the 2 hour search
>again and the results were a bit different.  The moves were the same (with
>similar times) but the summary block became:
>
> time=122:22  cpu=99%  mat=0  n=209963723  fh=89%  nps=28595
>              ext-> checks=21642959 recaps=582085 pawns=10695067 1rep=7462442
>              predicted=0  nodes=209963723  evals=34605063
>              endgame tablebase-> probes done=0  successful=0  maxd=32
>              hashing-> trans/ref=33%  pawn=10%  used=w99% b99%
>
>                               James B. Shearer


searches to the same _depth_ should be deterministic.  Searches to the same
_time_ won't be, although they should be close.

pawn hash hits at 10% is _very_ low, it just means you are over-searching the
pawn hash table badly which slows things way down (most pawn evaluations are
now computed rather than hashed).  I normally see 99% pawn hashes all the
time here...



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.