Author: Mike Byrne
Date: 16:41:53 11/20/03
Go up one level in this thread
On November 20, 2003 at 18:57:24, J. C. Boco wrote: >Doubling the processor speed or the time for thinking adds about 80 points to >the SSDF-computer-ratings. But PC's are already running so fast that doubling >their speed might mean searching at 15ply instead of 14ply. 80 points might be true for computer vs computer games - that will not be true for human vs computer IMO. For human vs computer , on the current top end machines , I say it would be less than 50. I think for computer vs humans on the low end units (say rated 2000), doubling might be worth about 75. I am sure many people will diagree with me. > >But dedicated computers are reaching, what, about 6ply on a search. If you >double the speed (or time) for a dedicated machine it might mean a difference of >7ply instead of 6ply. It seems to me that this would be much more of an >improvement. > >Is it fair to say that the 80points gains for doubling your speed is too low for >dedicated computers? I've been trying to come up with a logical (if >conjectureable) scale of the Star Diamond's playing strength per level, and the >number I've found useful is 150 points for each doubling. I totally disagree - but that is what makes the world interesting. Star Diamond is about 3x faster than Saphire 2. IMO, that may be worth about 100 points, I think the Sapphire 2 has about 2150 USCF strength. I would put the Diamond at 2250 -- perhaps it could be 2300 - but then maybe Sapphire 2 is perhaps 2200. One is for sure, the Star Diamond is currently the strongest unit available today. Only a few machine are potentially stronger - TASC R30/40, Genius 68030 etc. I am not certain , perhaps SSDF could test the Star Diamond against these older units. > >Just as a gross example of my little thought experiment (and I'm only doing this >because I have to wait until monday to get my computer!). > >Let's say the Star Diamond is rated 2200 at 40/2. 2200 is just the number I'm >using, since I'd like to think the $270 computer I just bought is a master. If >I assume that with a dedicated computer each doubling is worth 150 points >(instead of 80) then I get, with some aestetic rounding: > >Average time per move "Rating" > >3 minutes 2200 >2 minutes 2100 >1 minute 1950 >30 seconds 1700 >15 seconds 1550 >10 seconds 1450 >5 seconds 1300 >2 seconds 1150 > >While this loosely follows the Log(speedup)/Log2 * 150, it seems to me the >lower levels may be rated to low. I mean, I would think the 5 seconds of >thought for the latest dedicated computer would be stronger than mid-class D. > >If I used 100 per doubling then level *2 seconds* would be about 1500, which >strikes me as being too high. > >Perhaps 125 per doubling is closest. This would yield the lowest level a rating >of 1325. Hmmmm, yes I'm working it out as I'm writing. Maybe 125 ELO per >doubling is appropriate for dedicated chess computers. To me, the math is not working - try 75 and work your way down. 3 minutes 2200 2 minutes 2150 1 minute 2075 30 seconds 2000 15 seconds 1925 10 seconds 1875 5 seconds 1800 2 seconds 1725 I think this is reasonable for a dedicated unit rated @2200. > >Anyway, I'd be interested in the thoughts of others who most likely have thought >more about these kinds of things, or are in a position to know. Perhaps you've >played this computer yourself? > >And what is the latest commonly accepted formula for converting the SSDF >computer ratings to human USCF ratings? Is it still to add 180? Not since USCF took the Adult ratings down over a hundred points ove the last several years. Perhaps add 50 - I would add "0".
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.