Author: Matthew Hull
Date: 17:49:38 09/06/02
Go up one level in this thread
On September 06, 2002 at 20:14:52, Uri Blass wrote: >On September 06, 2002 at 19:53:52, martin fierz wrote: > >>On September 06, 2002 at 19:18:20, Matthew Hull wrote: >> >>>On September 06, 2002 at 18:04:29, Uri Blass wrote: >>> >>>>On September 06, 2002 at 18:01:11, Uri Blass wrote: >>>> >>>>>On September 06, 2002 at 17:32:47, Matthew Hull wrote: >>>>> >>>>>>On September 06, 2002 at 16:13:40, Robert Hyatt wrote: >>>>>> >>>>>>>On September 06, 2002 at 16:04:58, ALI MIRAFZALI wrote: >>>>>>> >>>>>>>>From the Threads here I am assuming that professor Hyatt beleives that 100X >>>>>>>>factor in speed (NPS) would be too much to overcome with software improvement >>>>>>>>factor.I am proposing the following possible match:Time control 40/2 6 >>>>>>>>games : GNU chess 5.04 on a pentium 4 at 2.4 Gigahertz vs Chessmaster2 original >>>>>>>>playstation (33 Mhz).This is actually a 73 factor in terms of processor speed >>>>>>>>which is not 100 but close.On the original playstation Chessmaster2 gets about >>>>>>>>1100 Nps. >>>>>>> >>>>>>> >>>>>>>Why gnuchess? I don't know much about it, and it might be perfectly ok. >>>>>>> >>>>>>>But you are also misinterpreting what I said. I did say that a factor of >>>>>>>100x, between programs that are "close" is overwhelming. Obviously a bad >>>>>>>program at 100X will be better, but it might not be much better. >>>>>>> >>>>>>>In any case, give your test a go and see what happens first... >>>>>> >>>>>>I'm running a test now with gnuchess (900mhz Duron) versus Crafty18.15 (90mhz >>>>>>Pentium). Gnuchess runs 16x faster on the Duron than the P90. At 40/30min >>>>>>minutes and after 36 games, gnuchess is 52% against crafty (not too impressive >>>>>>for gnuchess). The lower the time control, the better gnuchess does, of course. >>>>>> I have lots more data at home on this test, as well as an equal hardware test. >>>>>>I'm trying to get at least 40 games in each category, including 40/120. >>>>>> >>>>>>Not sure if the test will prove useful, but I'm thinking that one can do this >>>>>>experiment with any two engines and derive a function with which to calculate >>>>>>the speed advantage needed to reach parity/superiority by the weaker engine, >>>>>>qualitative factors aside. >>>>> >>>>>Thanks for your tests. >>>>> >>>>>I am interested to know how much games gnuchess lost on time because based on my >>>>>experience gnuchess lose minority of it's games on time at x minute/y moves. >>>>> >>>>>It may be more interesting to use fisher time control because I believe that gnu >>>>>chess does not lose on time at fisher time control. >>>>> >>>>>I still expect gnu chess to lose at slow time control inspite of the hardware >>>>>advantage even at fisher time control like 150 minute per game+25 second per >>>>>move but it is only a guess. >>>>> >>>>>I suggest that you use 6+1,30+5,150+25 as your 3 categories of time control. >>>>> >>>>>Uri >>>> >>>>Note that my experience is based only on games with no pondering and it is >>>>possible that things are different with pondering(I do not know). >>>> >>>>Uri >>> >>> >>>Pondering does not help the time problem. >>> >>>Also, weaker chess programs will perform better at shorter time controls, and >>>worse at slow time controls.:-) >>> >>>Gnuchess results so far with a 16x speed advantage over Crafty on p90: >>> >>>40/5 71% >>>40/10 53% >>>40/30 50% >>> >>>Of course, if Crafty were on the 900mhz and gnuchess were given a 16x time >>>advantage, the numbers would be lower. The speed advantage buys you less at the >>>greater depths I think. >> >>that's interesting! wasn't berliner's hitech/lotech experiment trying to find >>exactly this behavior, but it failed? > >No >There was no proof for diminishing returns but this experiment is different. > >> i made a similar experiment with my >>checkers program, but the "dumb" version of cake with a deeper search >>consistently beat the "intelligent version" with a shallower search over a wide >>range of search depths. > >One of the reasons that weaker programs are weaker is that they use inferior >search rules. >I believe that gnuchess use inferior search rules relative to crafty. > >In your experiment the dumb version had no inferior search rules. > >Uri Right. And if I do the same experiment replacing gnuchess with Phalanx, it will make a different "curve", since Phalanx is much stronger than gnuchess, but still weaker than Crafty, qualitatively (eval/search).
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.