Author: ShaktiFire
Date: 08:50:35 05/22/00
Go up one level in this thread
On May 22, 2000 at 06:18:47, blass uri wrote: >On May 22, 2000 at 06:11:59, blass uri wrote: > >>On May 22, 2000 at 04:31:58, Jouni Uski wrote: >> >>>On May 22, 2000 at 03:34:28, blass uri wrote: >>> >>>>On May 22, 2000 at 01:38:12, Jouni Uski wrote: >>>> >>>>>On May 22, 2000 at 01:27:53, blass uri wrote: >>>>>> >>>>>>They do not indicate it because you do not know what is the rating improvement >>>>>>from 4 processors relative to single 450Mhz. >>>>>> >>>>>But I know, it's at least 100 points as comp-comp games have indicated! >>>>> >>>>>Jouni >>>> >>>>You cannot know because the difference in comp-comp games does not have to be >>>>the same as comp-human games and I am also not sure that the difference in >>>>comp-comp games is 100 points because the ssdf did not test the new hardware. >>>> >>>>We can know the difference between p200 and K6-450 but we cannot know the >>>>difference between k6-450 and better hardware because it is possible that there >>>>is diminishing returns from hardware. >>>> >>> >>>Hasn't Heinz new book conclusion: there is no proof about diminishing returns >>>so far... >>> >>>Jouni >> >>There is no proof so we do not know if there is diminishing return or there is >>not. >> >>Uri > >The only way to prove that there is not diminishing return is by games and there >are not many comp-comp games at tournament time control to compare. > >We know that there is no significant diminishing return when you upgrade from >p200 to K6-450 but we do not know what happens when we upgrade from K6-450 to >better hardware. > >Uri The evidence is there in the ratings. At one time, SSDF correlated with human-computer games. Now the SSDF rating is much higher than actual human- computer results. This indicates..... against humans, there is a diminishing return with increase in computer speed. In computer - computer ratings, the ratings are still increasing linearly with ply. Its hard to plot this out, with different software versions, but the trend from p90 to p 200 to k6-450 is there... basically linear ratings increase with ply increase. This might change later, since I believe the actual curve, ratings vs ply in computer vs. computer, is piecewise linear, and gradually decreasing, but not decreasing as quickly as in comp vs human play.
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.