Author: Robert Hyatt
Date: 07:44:21 08/30/01
Go up one level in this thread
On August 29, 2001 at 23:08:30, Uri Blass wrote: >On August 29, 2001 at 22:14:28, Robert Hyatt wrote: > >>On August 29, 2001 at 17:55:40, Alberto Rezza wrote: >> >>>On August 29, 2001 at 10:41:45, Robert Hyatt wrote: >>> >>>>So the 1992 (roughly) version of the program, made using 3 micron ASICS, >>>>was at _least_ as strong as either deep fritz or deep junior on today's >>>>hardware. I will be conservative and say they are "equal". >>> >>>Of all the things that have been said in this thread, I find THIS the hardest to >>>believe. 1M n/s is not that much. And that was without null move, right? >>>How much shallower was their search? 5 plies less than DF/DJ, perhaps? >>> >> >>I don't know. I saw DT search 10-11 plies deep. I do about 2 plies deeper >>than that with null-move R=2. I am not sure how deep Fritz searches, but >>unless it uses some additional pruning, I would expect 12-13 just like me >>at 1M nodes per second. > >From my experience Deep Fritz can usually search deeper than 12-13 plies and >depth 15 is more realistic for tournament time control. OK... then he is doing some sort of forward pruning besides simple null-move. > >> >>Remember that a null-move search goes deeper, but experiences failures that >>"hide" things when you drop 2 plies here and there. It is not free. >> >> >> >> >>>And the eval was probably nothing special either - this is DT 1, not DB 2. >>> >>>I'd say it would have been very lucky to score 15% against DF or DJ... >> >> >>I disagree. I played some games with Cray Blitz vs Crafty. CB was >>consistently searching 1-2 plies less deeply (it had hardware about 7x faster >>than Crafty in terms of NPS). Yet it won the 10 game match convincingly. > >I believe that the main problem is that the time control was too fast. >I also believe that the evaluation of Crafty is better than Cray blitz because >tuning the evaluation to the right parameters is more important than the >knowledge that the Cray has and Crafty has not. > >Uri The evaluations are not that different, except that Cray Blitz still has a few expensive things in it that I can't do due to the computational cost. It had a _really_ good mobility algorithm that did way more than just count the squares a piece could safely move to. It had qualitative weights for each square so mobility to good/important squares was worth more than (say) mobility to a8. I simply can't do that on a PC, the software implementation of the vector merge instruction would simply be too slow to bear. With the exception of the null-move R=2/3 stuff in crafty, and the q-search checks and threats, plus singular extensions in Cray blitz, the two programs are very similar, IMHO.
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.