Author: Dann Corbit
Date: 14:27:04 01/27/00
Go up one level in this thread
On January 27, 2000 at 17:13:34, Timothy J. Frohlick wrote: [snip] >Right on Tim, > >This is especially true when the 1 and 2 ply searches may be full-width and the >30 and 31 ply searches are always selective searches. We know that DB was >thousands of times faster than any chess computers in the average home and it >only managed to hold its' own against Kasparov. If the rating goes up linearly >then DB should have had a rating of elo 3000+++ > >A speeding bullet is only as good as the person aiming it. Current estimates are about 50 ELO per doubling in speed. Assuming that Deep Blue's evaluation was the same quality as a PC (some say it was not that good, but it makes an interesting starting point), and we will also start with the assumption that the NPS of a PC an rating have some nearly linear correspondence, consider the following: Imagine a slowed-down Deep Blue that does 195,312 NPS and has an ELO of 2400 which would be roughly commensurate with many of todays PC programs. 200M is (coincidentally ;-)) exactly 1024 times faster. That's ten doublings. That would be an ELO of 2900, which is less than 100 ELO from the measured TPR for the second match. Well within expectations, wouldn't you say? An ELO of 3000 would require two more doublings -- meaning that Deep Blue would have to perform 800M NPS. So your math is off by a factor of four (roughly).
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.