Author: Robert Hyatt
Date: 12:30:07 09/01/99
Go up one level in this thread
On September 01, 1999 at 14:31:51, blass uri wrote: >On September 01, 1999 at 14:20:20, Robert Hyatt wrote: > >>On September 01, 1999 at 14:01:37, blass uri wrote: >> >>>On September 01, 1999 at 09:35:57, Robert Hyatt wrote: >>> >>>>On September 01, 1999 at 08:57:57, blass uri wrote: >>>> >>>>>On September 01, 1999 at 07:47:20, Robert Hyatt wrote: >>>>> >>>>>>On September 01, 1999 at 06:11:23, odell hall wrote: >>>>>> >>>>>>>Hi >>>>>>> >>>>>>> >>>>>>> Is there a huge difference in terms of elo points between rebel 10 on the >>>>>>>K6-600 amd and what my system K6-II350 ? The rebel page suggest that the >>>>>>>diffence is very minor, 2554 for k6-amd and 2576 for k6600? The reason I am >>>>>>>asking is because i am getting a little obssesed with the need to have the >>>>>>>strongest program, even though I could not beat rebel on my 486!!! I am >>>>>>>wondering if it would really be worth it to upgrade? What's a few rating points? >>>>>>>I probally would not notice the difference anyway. I notice also that my rebel >>>>>>>found all the moves of schroeders Amdk6 600 in the last grandmaster challenge. >>>>>> >>>>>> >>>>>>As I have said before, if you double the speed, your 2x faster rebel will >>>>>>be about 70 points better than the slower one, when they play each other. >>>>>>Against humans, there is too little data but it is certain that the 2x faster >>>>>>one won't be 70 points better than humans it was playing equal with at the >>>>>>slower speed... >>>>> >>>>>We have information that it is the case about p200MMX vs p90 >>>>>but we do not have information if it is the same at tournament time control for >>>>>faster hardware. >>>>> >>>>>The only way to prove that it is the case or to prove the opposite is by playing >>>>>games with faster hardware. >>>>> >>>>>Uri >>>> >>>> >>>>Actually we do have this. The first such experiment was run on horribly slow >>>>hardware years ago. Then Ken Thompson did the same with his belle machine and >>>>got the same results although at least 100X faster hardware was used. Hitech >>>>did the same thing for longer time controls. All three agreed. Monty Newborn >>>>and Ernst both investigated searching way deeper which simulates faster hardware >>>>than anything other than Deep Blue has, and they _both_ found that deeper is >>>>still better... with no 'tapering off' at least thru ply 15 in the middlegame. >>>> >>>>I'm convinced that faster = better, and that it is still linear and not getting >>>>worse. >>> >>>I am not convinced that it is still linear. >>>The only proof can be by games and not by the probability to change your mind or >>>to find a new move because it is possible that the new move does not change the >>>result. >>> >>>There is a point when it stop to be linear because you cannot improve forever >>>and it is not clear where is this point. >>> >>>I know results of many games of pentium200 vs p90 at tournament time control but >>>I do not know the same for faster hardware(p450 vs p200). >>> >>>Uri >> >> >>The first examples I gave _were_ from games. First using 'tinkerbelle', then >>using belle, then using hitech. All three produced similar win/lose percentages >>over shallower/slower searches, even though the games played with tinkerbelle >>and then belle didn't go as deeply as later games... > >The question is if the level in the first examples was better than p200 vs p90. > >Uri No... but the last one was... ie hitech was about as fast as a good P200 program today, probably faster (175K nps, plus it had hardware eval so it did a good bit of eval work in parallel as well...) I don't claim that things are linear. I do believe that they are, as so far, there has been no evidence to the contrary to show that an additional ply returns less and less as the search goes deeper...
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.