Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Endgame position, hash and EGTB cache

Author: Dieter Buerssner

Date: 06:52:51 01/24/01

Go up one level in this thread


On January 23, 2001 at 15:45:13, Ernst Walet wrote:

>I'm wondering what gives better results, conservative probing with 3-4-5 piece
>tb's, or agressively probing without the 5 piece.  The latter hardly slows down
>the search due to the fact that the 30 MB of 3 and 4 piece tb's easely fit into
>the cache (if large enough of course).  What do you think, Dieter?

My own experience is only from a subset of 5-men TBs. I have KPPKP,
KRPKR, KRNKR and KRBKR. Accidently, I believe, that the former two are
the most important. For the position tested by Dann (and by me with inferior
hardware) aggressive probing seems to help, despite the fact, that
the search speed in nodes/sec can get much less. The time to reach a
certain depth will show a slightly different behaviour, because of cutoffs
in the search tree due to TB hits. And even at the same depth, the analysis
with probing during the whole search should be better (at least not worse)
compared to the analysis with probing during only the upper parts of the
search tree.

You are right, with only 4-men, I get almost no decrease in search speed.
And there were positions I tested, that were solved faster, with only 4-men.
But I could not find a pattern. Sometimes also some luck is involved. You may
probe many positions close to the leaves of the search tree, which
the eval of the program will easily detect as won or lost. In other positions
this may not be that easy for the eval (i.e. KRPKR without any special knowledge
of this).

-- Dieter




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.