Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: DB NPS (anyone know the position used)?

Author: Peter W. Gillgasch

Date: 05:57:14 01/26/00

Go up one level in this thread


On January 26, 2000 at 03:14:11, Tom Kerrigan wrote:

>On January 25, 2000 at 23:38:06, Peter W. Gillgasch wrote:
>
>>Ad 1: This was true for Deep Thought since the difference
>>      between the fast and the slow eval was noticeable
>>      due to the sequential properties introduced by
>>      the Xylinx devices looking at the position on a
>>      file by file basis. Since DBII is not hampered
>>      by issues like FPGA capacities this is the first
>>      bottleneck that was to be removed. As Dave Footland
>>      has reported they had interesting things in their
>>      evaluation. In the light of this and in the light
>>      of things Hans has said it it extremely unlikely
>>      that they ever take a "slow eval" in DB since there
>>      is (a) probably no speed gain in doing so and (b)
>>      things like pins and king safety can add up quite
>>      a bit, taking a "slow eval" makes no sense in a
>>      machine which knows that the queen is pinned and
>>      will be lost versus a bishop or that there is a
>>      mate in one versus your king.
>
>I'm not sure I follow this. It seems like both a fast eval and a slow eval would
>make sense, depending on the situation.

This is an assumption from the "it is a tree search problem,
we need more speed and another full width ply please and the
tree search will see one ply later that I am going to be
checkmated, so why should I care" department.

Feng did design DT for speed, DB for more speed (still with
cramped pieces all over the place) and finally DBII for more
evaluation. As soon as you introduce a slow evaluation you
introduce the usual  inaccuracies and if you have pin detection
__hardware__ why should you be so idiotic and believe that since
your run off the mill equivalent of a "fast eval" (say piece
placement and material) is 4 pawns outside of the A-B window
that it is "safe" to elect to ignore the equivalent of a software
"slow eval" ? In the first DB - GK match, how many points does
a run off the mill "fast eval" assign to the black rook on h8
and the black bishop on b8 ? 8 pawn units, probably. Jesus
Christ ! That stuff was worth __zero__. If Feng didnīt understand
that he needs to fix that for real after that game, well, I
donīt know.

I suspect that there are other guys reading this forum
who are in a situation where there is no gain speedwise in
electing to ignore the "slow" evaluation. Testing for taking
the slow evaluation only is is my case more expensive than
taking the full evaluation (4 clock cycles). It helps :)

>I read somewhere that DB's fast eval took 3 cycles and the slow eval took 11
>cycles. I'm not 100% sure about the 11 cycles, though. It may be 12.

I havenīt seen any clock cycle descriptions of DB. You are probably
mixing things up with DT unless you can specify "somewhere" ;)

-- Peter



This page took 0.02 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.