Author: Graham Banks
Date: 23:43:06 02/27/06
Go up one level in this thread
On February 28, 2006 at 02:21:41, Graham Banks wrote: >On February 28, 2006 at 02:17:11, Ryan B. wrote: > >>On February 28, 2006 at 02:08:03, Graham Banks wrote: >> >>>On February 28, 2006 at 01:59:07, Ryan B. wrote: >>> >>>>On February 27, 2006 at 23:51:54, Uri Blass wrote: >>>> >>>>>On February 27, 2006 at 22:25:44, Ryan B. wrote: >>>>> >>>>>>On February 27, 2006 at 21:55:05, Charles Wane wrote: >>>>>> >>>>>>>I just got finished playing a four game match at game/30 min between Hiarcs 10 >>>>>>>and Rykba 13d, Hiarcs had an Athlon64 3200 with 128mb hash, against Rykba >>>>>>>running on a Pent II 450 with 128mb hash. The score was 2-2 !! Yesterday I >>>>>>>played about 14 games between rykba running on the same 450 against Deep fritz 8 >>>>>>>on the Athlon64 and rykba won the majority of the games!! Iv'e never seen a >>>>>>>program with such inferior hardware play so strong. I actually rated rybka on >>>>>>>the 450 on chessmaster 10th but letting rybka play the various personalities it >>>>>>>ended up with a rating of 3055!! And lost only one game , running all on a mere >>>>>>>Pent II 450. I guess Rybka's case the chess knowledge means much more then the >>>>>>>speed of the program? I don't know cause I'm not a computer chess expert, but >>>>>>>all this seems bizarre. >>>>>> >>>>>>From what I have seen Rybka has great knowledge but not in the way of >>>>>>traditional chess knowledge by having a huge evaluation function. Like all >>>>>>chess engines I expect Rybka to play better on faster hardware but I so far >>>>>>think the gain of fast hardware is less than most other programs. From what I >>>>>>see Rybka ranks higher in blitz than long tc games. This is at least the case >>>>>>on the CEGT rating list. >>>>>> >>>>>>Ryan >>>>> >>>>>I do not think that you are right here and rybka does not rank higher in blitz. >>>>> >>>>>Let compare the 32 bit versions of rybka >>>>> >>>>>blitz >>>>>4 Rybka 1.01 Beta 13b 32-bit 2820 20 31 652 69.8 % 2675 23.0 % >>>>>5 Rybka 1.0 Beta 32-bit 2811 9 12 3139 68.1 % 2679 29.2 % >>>>> >>>>>long time control >>>>> >>>>>3 Rybka 1.01 Beta 13-13b 32-bit 2869 28 28 494 73.9 % 2689 27.9 % >>>>>5 Rybka 1.0 Beta 32-bit 2823 10 10 3670 68.8 % 2686 31.1 % >>>>> >>>>>You can see that rybka has higher ranking at long time control. >>>>> >>>>>I believe that the 64 bit version simply score too much to have reliable rating >>>>>because it played in the long time control against opponents that are too >>>>>weak(more than 300 elo weaker than it). >>>>> >>>>>Uri >>>> >>>> >>>>Fruit is my reference point. >>>> >>>>CEGT 40/40 >>>>Rybka 1.0 Beta 32bit 2816 >>>>Fruit 2.2.1 2779 >>>> >>>> >>>>CEGT Blitz 40/4 >>>>Rybka 1.01 Beta 13b 32-bit 2820 >>>>Fruit 2.2.1 2767 >>>> >>>>Ryan >>> >>> >>>CCRL 40/40 (Athlon XP3800+) >>>Rybka 1.0 Beta 32 bit 2883 >>>Fruit 2.2.1 2848 >>> >>>For all intents and purposes the same points differential as the CEGT 40/40, so >>>the gap doesn't seem to decrease much at longer time controls. >>> >>>Regards, Graham. >> >> >>Interesting, thanks for showing me that. What is the web site for CCRL? >> >>Thanks, >>Ryan > > >http://computerchess.org.uk/ccrl/4040/ It's actually based on 40/40 Athlon 64 3800+ An error on my part. Graham.
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.