Author: Howard Exner
Date: 13:56:00 10/10/00
Go up one level in this thread
On October 10, 2000 at 12:51:23, Uri Blass wrote: >On October 10, 2000 at 12:42:11, Howard Exner wrote: > >>On October 10, 2000 at 12:38:19, Uri Blass wrote: >> >>>On October 10, 2000 at 12:26:14, Howard Exner wrote: >>> >>>>"Looks like a doubling of NPS no longer provides an extra 50 Elo rating against >>>>humans - nothing even close, in fact." - g >>>> >>>>I've never subscribed to that for comp vs human. Many have proposed that the +50 >>>>or so gain is valid for comp vs comp. That I can live with. For humans maybe >>>>it's closer to 15-20 for doubling? >>> >>>For comp-comp it seems to be at least +70 for doubling based on the ssdf >>>results. >>>For comp-humans I believe it is at least +30. >>> >>>Uri >> >>Is that for just a speed double, or a speed double plus a new program version? > >Just for a speed double. >You can see the rating of some programs like Junior5,Fritz5.32,Hiarcs7.32. > >I remember that the difference between P200 and K6-450 was 70 elo rating and >even slightly more than it. > >The K6-450 is twice faster(not exactly but there was not a big difference) > >Uri The reason I'm guessing around 15-20, about 1/4 or 1/3 of the comp vs comp effect, for strong GM's is I'm not convinced a mere speed doubling will address their pattern recognition/ knowledge which allows them the opportunity to come up with deep winning plans. The doubling of the computer will double what computers do best - calculate. Will it double their knowledge? Yes, in the sense of being given the opportunity to exploit that knowledge by hitting deeper positions. Speeding up calculating can resolve some positions, but those situations may not be enough to hamper the GM elite. An example is a strong knight vs a weak bishop ending. A GM or lesser player would devise a long range plan of swapping down knowing that the pattern is usually winning. A computer, 2x, 4x, or 8x may not have a clue what's going on without the knowledge. Sometimes programs even think they are winning, because they do not re-adjust the minor piece values depending on the position. How do we measure all this? - Would a half speed Junior make the same key moves as the Quad? - Calculating the speed doublings between Deep Blue (1996) and Deeper Blue,but taking into account Deeper Blue was also improved in the knowledge department also. Then asking Gary Kasparov what he felt the overall rating improvement was between the two. By the way, I am assuming a discussion on Strong GM's, as the original post referred to Junior's 2700 rating competing against such players( from another post of yours, I agree that Junior's results were impressive). I liked Graham's post as it provoked alot of thought in me.
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.