Author: Robert Hyatt
Date: 06:15:11 04/20/00
Go up one level in this thread
On April 20, 2000 at 01:16:44, Paulo Soares wrote: >On April 19, 2000 at 23:56:39, Robert Hyatt wrote: > >>On April 19, 2000 at 23:39:03, Paulo Soares wrote: >> >>>On April 19, 2000 at 23:31:13, Robert Hyatt wrote: >>> >>>>On April 19, 2000 at 23:19:22, Drazen Marovic wrote: >>>> >>>>> In around 3 years Bushinsky hypothesizes no human can win in an 8 game match. >>>>>Well If deep blue is rebuilt and upgraded then yes, oterwise no IMHO. PC comps >>>>>will be between 2500 and 2680 for the next 8 years IMHO. >>>>> >>>>>"Shay Bushinsky answers >>>>> >>>>> >>>>>The term unbeatable is a very tough word indeed. I cannot say that a computer >>>>>will ever be unbeatable in any given game in the near future, but I do see a >>>>>time - maybe in about three years from now, when a computer will always prevail >>>>>in an 8 game match against the strongest human " >>>>> >>>>>Taken from the kasparov site >>>> >>>> >>>>I also expect to see a pig flying in about 3 years, +/- 50 years. :) >>>> >>>> >>>>In 10 years, this _might_ be true. I don't think in 3 years however.. >>> >>>Why 10 years? >>> >>>Paulo Soares, from Brazil >> >>that is about how far we are behind the hardware speed of deep blue, based on >>classic Moore's law. It is only a wild guess... I would say 10 years at >>least... maybe more... and only if Moore's law holds... > >Moore's law: "To every 18 months the processors double the speed". In 10 >years=120 months, we will have 120/18=6 times in that the processors will >double the speed, what means that in 10 years we will have processors with >approximately 1Gz(actual frequency) x 2(6) = 64 Gz! >But about the software, you don't think the development of the software will >have a weight reasonable on these ten years? > >Paulo I was being a bit more speculative on Moore's law. For simplicity I use 12 months (one year) rather than 1.5 years... which means I think we need about 10 doublings = 1000x faster than today, to approximate what DB was doing... They were doing 200M nodes per sec to a max of 1B nodes per second (200M was quoted as 'nominal'). that factors into about 100x faster than today's very best speeds using parallel search. Their eval could also do about 10x the work ours do, all done in hardware in a few cycles. 100x * 10x = 1000x, which is the increase in speed I believe we need to approximate what they have done. 1000x means 10 years optimistically... or 15 years if you really stick to the 18 month doubling time. As far as software improvements, yes there will be some... but many of our software improvements are merely 'catch up' to what they were already doing in their evaluation... So I don't think it will affect the number at all...
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.