Author: Robert Hyatt
Date: 21:14:20 05/15/99
Go up one level in this thread
On May 14, 1999 at 13:14:36, KarinsDad wrote: >On May 14, 1999 at 05:24:20, Peter Hegger wrote: > >>Hello >>Let's say that today's best programs, Fritz, CM6000, junior etc.. are playing at >>the 2450 level at 40/2 when they've got hardware capable of knocking off .5M >>nps. I don't think this is too outlandish an assumption. >>If you double this speed 8 times over you arrive at 128M nps. This is in the >>same ballpark as this new proposed screamer of Hsu's which it is estimated will >>knock off 120M nps on a multi-processor platform. >>I've seen in other threads that doubling speed will increase performance >>anywhere from 30-70 points per doubling. For argument's sake and to split the >>difference I'll assume that 50 is likely pretty close. Using 2450 as the base >>this would translate into an elo of 2850 give or take a bit. >>Is it really possible that a machine which is stronger (marginally) rating wise >>than the world champion is right around the corner. Or am I missing something >>here in making this estimate? >>In any event I'd love to see Kasparov tackle this baby in a 40/2 24 game match. >>Bets anyone? :) >>Regards >>Peter > >As stated in other posts, the 30-70 points per doubling starts dropping at the >larger elos, so using conventional chess theory, this might result in a rating >of 2700 or so. > >However, consider the following: > >At 120 mnps, you could calculate ALL moves (i.e. exhaustive search) for most >positions to ply 6 in about 8 seconds. In a 40/2 game, this would leave you with >about 172 seconds per move to calculate normally (minmax) beyond ply 6. This >would mean that the program would never make a tactical mistake below ply 6. >This should be a major advantage over current max ply 2, 3, or 4 exhaustive >search program >Additionally, Garry Kasparov normally looks 6 to 10 ply into a position (at >least this is what he claims). This would mean that whereas Garry would often be >looking 6 ply selectively into a position, the program would be looking 6 ply >exhaustively into the position. you are forgetting alpha/beta, which returns the same score as minimax, but searches 2*sqrt(N) nodes to do so (N = total moves searched by minimax search, or roughly 38^D where D is the depth in plies of the search. In effect, alpha beta lets us search about 2x as deep as we would with pure minimax... Typical search depth for crafty in the opening/middlegame phases, using the quad xeon at 1-2 minutes per move is 12-13 plies full-width, using alpha/beta, PVS, null-move R=2, and the abbreviated q-search I use. > >Considering that at 40/2 the program would still have 172 seconds to consider >beyond ply 6 (on average per move), it becomes apparent that it would become >more and more difficult for Kasparov to come up with any sort of tactical >combination that the program could not foresee. you are overlooking the difference between 'average' and 'maximum'. I have seen _many_ GM players find forcing tactical lines that are > 30 plies deep, and they have worked it out before playing it. So they go _far_ deeper than 6 plies along forced lines. I do as well. IE if you can solve win at chess 2 yourself, you can search deeper than 6 plies... > >Also, the program could exhaustively search ply 6 and some portion of ply 7 >during pondering (on average) so that when it is it's turn to move, it would >usually have it's full 180 seconds to start working at ply 8. > >So, the bottom line of what I am saying is that a programmer could change his >paradigm on what is normally allowable in a chess program at today's speeds and >include elements which are not even considered today. He could exhaustively >search to ply 6, he could exhaustively search all checks, he may drop out null >moves in some circumstances (and find an occassional position where 2 in a row >for the same side was not sufficient), he may search deeper after the quiescence >search and find more sacrifices, etc. The ability to "tweak" the design are much >greater than at the slower speeds where there is a constant fight to only >examine worthwhile paths, etc. The ability to change how to design the program >could gain 100 to 200 elo as well and such a program could possibly get as high >as 2900 elo at 40/2. > >KarinsDad :) I didn't follow the "6 ply" example since I normally search (Crafty) 12-13 plies in 1-2 minutes...
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.