Author: Robert Hyatt
Date: 14:08:52 11/07/98
Go up one level in this thread
On November 07, 1998 at 17:04:22, Robert Hyatt wrote: >On November 07, 1998 at 16:25:20, Graham Laight wrote: > >>On November 06, 1998 at 20:41:05, Robert Hyatt wrote: >> >>>On November 06, 1998 at 18:48:39, odell hall wrote: >>> >>>>I am interested in knowing, if even by a wild guess when a program with the >>>>strength of deepblue will be available to the public? Five years? or maybe >>>>three? >>> >>>Here's the basic math... DB searches about 250M nodes per second. Factor >>>in its evaluation, which is probably at least 10x as complex as what is done >>>in the micros, so this is 2500M nodes per second equivalent. A good micro >>>today would do 250K nodes per second... 2500M/250K = 10,000... so we need >>>to get a micro up to 10,000 times faster. If you figure the current doubling >>>every 1.5 years, log2(10000)= 13-14. So 1.5 * 14 == 21 years, roughly. >>> >>>However, it is doubtful the doubling every 18 months is going to continue, >>>so this is easily a lower bound on the time-frame... If you want to factor in >>>new software improvements, this might get shaved a few years... but it isn't >>>going to happen in 3-5 period... >> >>I'm sorry, but I feel the urge to disagree. >> >>The evidence of the play in the 1997 tournament against Kasparov is not >>compelling enough to support this assertion. >> >>In game 1, it lost through disappointing play. >> >>In game 2, Gary resigned from a drawn position - a potential draw that some of >>the micro software would not have allowed. > >this is wrong. *NO* micro can find either of those draws... Ed posted the >analysis on his web site... A micro might play a different king move, but *not* >because it sees the draw... > >> >>In games 3, 4, and 5, DB looked beaten, but GK couldn't put it away. >> >>In game 6, GK messed up the opening, leading to his defeat. >> >>GK has previously been beaten by a 90 Mhz Pentium at G25 time setting. >> >>Since the GK tournament, DB has been cowering in fear - not being allowed to >>play another man or machine in public. >> >>No better commercial chess computers until 2019? >> >>Hmmmmmmmmmmmmmm. > > >I don't believe so, no. Based on 10+ years of experience in watching older >and slower versions of deep thought absolutely shred micro programs, and >factoring in the 100-fold improvement (at least) in the speed of DB over >the older Deep Thought, I'd think that there might not be a better commercial >program for even longer if my suspicion that doubling in speed every 18 months >turns out to be true.. I don't see how it can continue... and without that >performance boost, micros vs db would be totally hopeless... I also overlooked one other issue... you sort of casually brush-off the way DB played against Kasparov, yet we have seen no other program even approach such a result at standard time controls... And it wasn't a matter of "Kasparov not being able to put it away..." In at least two of those games there was no win possible because DB simply found the proper defense... And let me remind you of the results someone else posted months ago, where two different commercial programs were given the white side of game six against only an IM, and the commercial programs lost *all* the games... winning from the white side is *not* trivial... remember that black was up a whole piece. Everyone wants to write that game off, yet we don't seem to have another program that could win it from the white side. So the gap is still *very* wide...
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.