Author: Jens Kahlenberg
Date: 15:03:20 08/19/03
Go up one level in this thread
On August 19, 2003 at 16:22:58, Uri Blass wrote: >On August 19, 2003 at 15:02:11, Jens Kahlenberg wrote: > >>On August 19, 2003 at 13:36:12, Uri Blass wrote: >> >>>On August 19, 2003 at 12:14:04, Jens Kahlenberg wrote: >>> >>>>On August 19, 2003 at 09:21:11, Uri Blass wrote: >>>> >>>>>On August 18, 2003 at 14:47:26, Jens Kahlenberg wrote: >>>>> >>>>>>On August 18, 2003 at 01:27:55, Uri Blass wrote: >> >>[snip] >> >>>>>>>Fritz3 already got the IM norm on p90 at 120/40 time control. >>>>>>>even without considering software improvements >>>>>>>3 minutes per move on p90 is equivalent to 10 seconds per move with the hardware >>>>>>>of today >>>>>>> >>>>>>>if we assume that 10 seconds per move is 2400 than 2 seconds per move is above >>>>>>>2000 even if you assume 100 elo for doubling the speed and I did not consider >>>>>>>the fact that the human play worse when (s)he has less time to ponder and that >>>>>>>the computer play better when it has more time to ponder. >>>>>>> >>>>>>>My guess is that when you consider also the software improvement from fritz3 to >>>>>>>the top programs of today you get that the top programs of today can perform at >>>>>>>IM level if they are allowed to use only 2 seconds per move. >>>>>> >>>>>> >>>>>>Hi Uri, >>>>>> >>>>>>I don't think so. IMHO even FM-level should be enough to survive the middlegame >>>>>>generally with a not totally drawn position if top engines have only 2 seconds >>>>>>per move, because they can't get deep enough to outsearch their opponent in >>>>>>tactical issues. In the endgame the engines will have a hard time to even draw a >>>>>>drawn game at that rate of play. >>>>>> >>>>>>Regards, >>>>>>Jens >>>>> >>>>>Fritz3(p90) already did the IM norm in 1994. >>>>> >>>>>Now even if you assume no pondering of more than 2 seconds Fritz3(p90) 180 >>>>>seconds per move is equivalent to Fritz3(hardware that is 20 times faster) 9 >>>>>seconds per move >>>>>and Fritz3(9 seconds per move) is probably at similiar level to top programs of >>>>>today (2 seconds per move) >>>>> >>>>>Uri >>>> >>>> >>>>If i get your "calculation" right, Fritz3 on usual hardware (not those 8-16 >>>>processors SMP-machines) _must_ have the strength to compete with Kasparov at >>>>normal rate of play. Ergo: Top engines on SMP-hardware _must_ kill him?! >>> >>>No >>> >>>Fritz3 on p90 is at FM/IM level >>>Fritz3 on usual hardware is probably GM level(anti computer experts may beat it >>>but most players play their normal game when they play a computer). >> >> >>Sorry Uri, >> >>i didn't know that your exact calculation just holds for one direction. Shame on >>me because i estimated the stupid other direction: >> >>Fritz3 on p90 is IM (2350 ELO) at normal rate of play. And doubling the speed >>four times seduced me to the wrong assumption that Fritz3 on a 1.44 GHz machine >>should reach 2750 ELO ... enough to compete with Kasparov. > >I think that your assumption about the difference in elo from doubling the speed >is wrong against humans. And i think that all calculations based on an IM norm of Fritz3 in 1994 on a p90 are _very_ fuzzy. IMHO only practice can answer the question. > >> >>But now i fear that you must've been right as took the other direction and >>wrote: >> >>> Fritz3 already got the IM norm on p90 at 120/40 time control >>> control. Even without considering software improvements >>> 3 minutes per move on p90 is equivalent to 10 seconds per >>> move with the hardware of today >>> >>> if we assume that 10 seconds per move is 2400 than 2 seconds >>> per move is above 2000 even if you assume 100 elo for doubling >>> the speed >> >>Perhaps i'll never understand this difficult ELO algebra for computer chess :-( > >I said even if you assume 100 elo for doubling(my assumption is that 100 elo is >only upper bound and it can be less than it). > >Uri I even don't believe that a marvellous linear function is behind the curtain :-) Best regards, jaxx
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.