Author: Uri Blass
Date: 04:50:44 12/03/00
Go up one level in this thread
On December 03, 2000 at 05:18:18, walter irvin wrote: >On December 02, 2000 at 16:00:50, Uri Blass wrote: > >>On December 02, 2000 at 15:45:55, Georg Langrath wrote: >> >>>On December 02, 2000 at 14:58:45, Uri Blass wrote: >>> >>>>On December 02, 2000 at 13:33:07, walter irvin wrote: >>>> >>>>>On December 02, 2000 at 12:06:17, Georg Langrath wrote: >>>>> >>>>>>It is interesting how strong chesscomputers will be in future only depending on >>>>>>hardware. Couldn’t you get an idea if you let two computers play against each >>>>>>others with auto 232? One of the computer had usual time controls and the other >>>>>>ten times longer time. Then you see what ELO the computer got that had the long >>>>>>time controls. In that way you could foresee how strong computers could get, >>>>>>when they are ten times faster than today. >>>>>> >>>>>>Georg >>>>>the test you want to run would be better if ponder were left off .because if the >>>>>program that gets less time predicts the move then the fact that it gets less >>>>>time will be cancelled out .so your test would be good if ponder was off on both >>>>>programs . >>>> >>>>I think that you can do the test with ponder on and you do not need that one >>>>program will get more time. >>>> >>>>The only thing that you need is different computers when one of them is 10 times >>>>faster. >>>> >>>>You can do a match between programs on p100 and programs on p1000 when the time >>>>control is 20 hours/40 moves and get the results. >>>> >>>>Uri >>> >>>Yes a good idea but there is one question. Is the difference in ELO the same >>>between p100 and p1000 as between p1000 and p10000? With my idea you can examine >>>that. I think that the difference is bigger between p100 and p1000 than between >>>p1000 and p10000. But I am not sure. I would examine myself, but I don't have >>>two computers. >>> >>>Georg >> >>I assumed that the faster computer is only 10 times faster and that it is the >>only difference in both cases. >> >>If this assumption is correct then I see no problem. >> >>I suggested time control of 20 hours/40 moves in order to predict the result in >>2 hours/40 moves between p10000 and p1000. >> >>Uri >at first sight your idea seems perfect .then i thought about chess itself .i >believe that the search tree for a 1ghz computer gets so big that a 10 ghz >computer will not make much head way .so to be 1000% sure of this all a person >would have to do is use 1 computer with ponder off both programs . I prefer different computers with ponder on because normal games are with ponder on. give 1 program >3 min the other 30 mins per move .i think a good test would be cm6000 i can play >it vs itself at those kinds of odds . It may be interesting to know the results but a better test is by games between different programs because I am not sure if the difference in rating is the same when you play the program against itself. i still believe that a 10 ghz computer does >not play much better at slow chess than a 1 ghz computer,maybe 100 elo .between >a 100 mhz and a 1ghz is about 500 elo . I assume that 1ghz is exactly 10 times faster than 100 mhz and that 10 ghz is exactly 10 times faster than 1 ghz. I do not think that the difference between 100mhz and 1ghz is about 500 elo. Assuming 70 elo for doubling you get about 160 elo difference. I believe that the difference between 1ghz and 10 ghz is smaller because of diminishing returns and I guess 100 elo difference. Uri
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.