Author: Robert Hyatt
Date: 18:47:57 08/22/02
Go up one level in this thread
On August 22, 2002 at 05:31:40, Gian-Carlo Pascutto wrote: >On August 21, 2002 at 22:35:53, Robert Hyatt wrote: > >>On August 21, 2002 at 14:51:17, Gian-Carlo Pascutto wrote: >> >>>On August 21, 2002 at 14:44:37, Robert Hyatt wrote: >>> >>>>>Please look at page 5 very carefully 12.2 ply it says. >>>> >>>>Yes. And take the logs and add up the _first_ number from each iteration >>>>and divide by N. Guess what you get? 12.2... >>>> >>>>Which was the software part of their search. >>> >>>They certainly don't say the last. >>> >>>A bit further, you can see that a nominal depth of 12 compromised 8 ply >>>nominal in software. Add 4-5 hardware ply and you get 12-13 ply. >>> >>>8 ply in software, plus hardware plies. Not 12 ply in software plus >>>hardware plies. >>> >>>The paper is pretty conclusive that they were not searching 18 ply >>>nominally, but only 12 ply. They even literally wrote it. >>> >>>-- >>>GCP >> >> >>It doesn't read like what the team said, that is for sure. But there are two >>key pieces of data that I _personally_ know to be true: >> >>(1) deep thought searched 10-11 plies in 1989 at the WCCC in Alberta. It >>continued to search that deep thru the 1994 ACM event in Cape May, New Jersey. >>I was there. I watched their monitor as we sat around the table watching their >>games. So there is absolutely no doubt about that 10-11 ply number as I _know_ >>what was happening in those games. >> >>(2) deep blue was claimed to be 100X faster. That should turn into 4-5 plies >>at their EBF as shown in the logs... >> >>More than that I won't speculate on. But Cray Blitz was searching 9-10 plies >>at 400K nodes per second, and I _know_ DT was out-searching us at _least_ a >>ply, plus their singular extensions that we were not using in 1989. > >I think Uri already sufficiently addressed this question above. Loss of >efficiency, more extensions, and the bad branching factor all contribute to >a minimal gain. > >-- >GCP One small problem... There is no _evidence_ of a bad branching factor. In fact, there is evidence of a good one...
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.