Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Fritz5 and memory

Author: Komputer Korner

Date: 08:29:17 06/01/98

Go up one level in this thread


On June 01, 1998 at 10:32:51, Don Dailey wrote:

>On June 01, 1998 at 03:20:07, Komputer Korner wrote:
>
>>Surely the 7% speedup on doubling only works until the hash table is
>>large enough to accomodate the complete search for the longest move
>>think in a game. So based on having a large enough hash table to
>>accomodate this on a time control of 40/2, are we all agreed that the
>>maximum hash improvement speedup is about 30% ( reason is more
>>middlegame moves than endgame moves ) maximum speedup or around 25
>>points at today's ratings and speed. Is this reasonable? We need tests
>>to see if programs play 25 points weaker without hash tables.
>
>Yes, you are correct.
>
>The speedup seems to ALWAYS apply if the search is deep enough.
>Once your tables are larger than than the search can fill, making
>them bigger will not help (but they will help a deeper search.)
>
>I probably should have made that point clearer.


>
>I think most programs benefit a lot more than 30%.  I think our
>serial program is at least twice as fast with tables, at around
>3 minutes per move, maybe more like 3 times faster.  Of course
>in engames this is a little higher and in simple endings it can
>be huge.  Are these the numbers others get?   I'm assuming it
>is ok to count the extra improvement of move ordering that the
>hash table suggestion gives you.
>
>- Don

If this is correct, then almost all programs should see a rating point
gain of between 60-90 points. We need tests with large hash tables and
without any hash tables.
--
Komputer Korner



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.