Author: Peter Kappler
Date: 21:31:06 09/05/00
Go up one level in this thread
On September 05, 2000 at 20:44:18, Jonathan Lee wrote: >On September 05, 2000 at 16:04:59, Peter Kappler wrote: > >>On September 05, 2000 at 15:47:33, Jonathan Lee wrote: >> >>> Statistically speaking (although imperfect and controversial) does doubling >>>the computer speed means adding 30 rating points for FIDE or SSDF? >>> I know people will say more or less, depends on time control, and it's not >>>linear at all. >>> Hopefully the top notch software have similar variables to each other. >>> Jonathan (64th message) >> >> >>Everything I've seen says 70 points per doubling. >> >>The question is whether or not this trend continues indefinitely. My personal >>opinion is that it must gradually taper off with increasing depth. >> >>--Pete >Allow me to be more specific: > > Let's make the time control 40 moves in 2 hours and sudden death 2 more >hours with human vs. machine. >Humans also have limits or tapering off on the middle game too. >Assume you have a GM database, how many GHZ will it take to match the middle >game move on a given position of both players who are top 10 in the world? > > Upon using the fastest PC (currently 1 GHZ), you can match the middle game >position and move "in a number of hours". >Since 40 moves in 2 hours equals 3 minutes per move, in 24 hours when the move >has a match, it means a 480 GHZ computer could equal the IGM. >24 hours divided by 3 minutes equals 480 on a 1 GHZ computer. >2^9=512 which almost equals 480 GHZ. 512==480 >9 times 70 = 630 rating points >I agree it MUST taper off; about 600 rating points is way too much. >FIDE rating 2500 + 600 = out of bounds > > Of course, GM knows that closed pawn structure and queens on the board adds >complexity and the best way to beat the machine. Disregarding complexity, at >some GHZ the computer reaches a 2800, 2850, and 2900 FIDE ratings under >tournament time controls. > That is my question. You could use Fritz 6 or the other close contenders. > We could also compare Deeper Blue moves (positions) with the current PC also >as long it takes about 24 hours on 1 GHZ. > Jonathan (65th message) The only way to answer your question is by getting a bunch of GMs to play slow time control games against computers. You can take Junior's recent performance in Dortmund as a reference point. It had a performance rating of 2700 on a machine with an effective speed of ~4GHz. I doubt that the 70 points per doubling rule holds up in comp vs human games. Extra speed is probably less important against humans. --Peter
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.