Author: Jeremiah Penery
Date: 04:05:49 10/10/02
Go up one level in this thread
On October 10, 2002 at 04:26:50, Gian-Carlo Pascutto wrote: >On October 09, 2002 at 14:15:10, Jeremiah Penery wrote: > >>On October 09, 2002 at 07:21:45, Gian-Carlo Pascutto wrote: >> >>>Please read http://sjeng.org/ftp/deepblue.pdf, written by the DB team, >>>which directly contradicts the 12(8) = 12+8 hypothesis. >> >>Could you please tell me where it contradicts this? I see "A three minute >>search on Deep Blue would reach a full-width depth of 12.2 on average." That >is not contradictory, as they appear to commonly refer to the full-width depth >>as only the number searched by software. >> >>There are a couple tables that you could be referring to, but they can't >explain what 12(8) really means, if it is not software(hardware) depth. And >>how can we explain searches like 4(5) if it works differently? > >I'm referring to the tables on page 13 and 14, that list their nominal >depth and the corresponding nominal software depth. If you use null-move R=3, your minimum software depth for iteration i will be at least i-3. Does this mean I can say you're searching 9 ply when you report 12 plies of output? The table lists iteration depth (which, from other examples, DOES NOT generally include hardware depth when counted by the DB team) and the corresponding MINIMUM and MAXIMUM software depth - NOT the 'nominal' depth. >The text on page 13 also states: > >'For a given iteration i, the software is assigned i-4, ...' To complete the quote: '... which represents the minimum depth search in software.' See above, the null move question. >and especially > >'When hardware search extensions and quiescence search are taken into >account, we typically see searches from 6 to 16 ply.' To finish again: 'Thus we can see iteration 12 searches can reach as deep as 40 ply..." If you see by their table, the maximum software depth "is approximately three times the minimum depth." So at iteration 12, minumum depth was 8, maximum software depth of 24 (23/24 in the tables). The '6 to 16 ply' thing was maximum hardware depth, which would include the extensions and quiescense they did (as per the quote). Maximum software depth 24 plus maximum hardware depth 16 gives you the quoted 40 plies. >Considering that the majority of positions does get an extensions or >ends up in qsearch, this makes it impossible that they were going 16 >ply nominally, since then they would average far higher than that. The >number makes perfect sense for 12-13 ply nominal searches, though. The 16 plies was maximum hardware depth, and had nothing to do with the software depth, iteration depth reported, or the "nominal plies" of search.
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.