Форум » » RobboLito vs Rybka 3 40/40 » Ответить
RobboLito vs Rybka 3 40/40
Azaad: RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off 1. Silver Opening Suite , 31 out of 100 games played 19.0 - 12.0 in favor of RobboLito ( +9 -2 =20 ) 2. Noomen 2008 opening Suite , 33 out of 60 games played 18.5 - 14.5 in favor of RobboLito ( +8 -4 =21 ) PGN after the matches are finished . Please note that the Silver suite match is played on a 3 GHz Core 2 machine . So my time control on that machine is equivalent to 40/60 CEGT and CCRL after hardware adjustment . The Noomen match is on a slower laptop which has same speed as CEGT and CCRL benchmark machine .
Ответов - 11 новых
Trini: Azaad пишет: RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off 1. Silver Opening Suite , 31 out of 100 games played 19.0 - 12.0in favor of RobboLito ( +9 -2 =20 ) 2. Noomen 2008 opening Suite , 33 out of 60 games played 18.5 - 14.5 in favor of RobboLito ( +8 -4 =21 ) PGN after the matches are finished . Please note that the Silver suite match is played on a 3 GHz Core 2 machine . So my time control on that machine is equivalent to 40/60 CEGT and CCRL after hardware adjustment . The Noomen match is on a slower laptop which has same speed as CEGT and CCRL benchmark machine . Greetings Azaad, More fuel to the fire.... I suppose that the results attacks a view that at longer timecontrols Rybka is better. Further results would be interesting. Peace.
Azaad: Trini The progress chart so far Stage 1. I don't believe that I/RL is better than R3 . Post some convincing results . Stage 2. OK , perhaps I/RL seems to be better than R3 but it's just an illusion because R3 has non-zero contempt play as default which causes it to lose drawn games . Stage 3. OK , but I/RL cannot ponder . Only ponder on games are real games . Stage 4. Even if I/RL can ponder and beats Rybka in ponder on games it's only as strong as Rybka 3 in long time control games . Stage 5 . I don't care if RL beats R3 1t in long time control games . Why are you crippling Rybka by running it on single core only ? Conveniently forgetting that - R3 is sold also as a SP engine so a prospective R3 SP customer needs to be informed that now his money would not be purchasing the strongest SP engine . - RL is as strong as R3 2t . So R3 has advantage over RL only on quads and above which are still a minority hardware . - R3 2t is still not as good as RL because on a dual core RL gives me same quality as R3 2t and leaves one core free on which I can still do other work . I , for one , never run engine matches simultaneously on all cores of any of my machines ( excepting the old resurrected P IV where I have no choice ) . It all reminds me of an old folk tale . The village idiot who could not count properly once had a bet with another villager about how many legs a goat has . The idiot said 3 , the other person said 4 . When the idiot's wife came to know of the bet she said ,"You fool , you've lost . A goat has four legs" . The idiot replied , "Who has lost ? Whatever proof he offers I'll just keep on repeating that a goat has three legs ! " Seems like the technique is still in widespread use .
никел: Uhh? Around +60 elo. Great Stage 6. OK , perhaps I/RL seems to be better than R3 at long time controls, but the elo difference is lower than at ultrafast time controls Stage 7. The comrades are liars, they said +50, and in fact RL is stronger than +50
Trini: Azaad пишет: Trini The progress chart so far Stage 1. I don't believe that I/RL is better than R3 . Post some convincing results . Stage 2. OK , perhaps I/RL seems to be better than R3 but it's just an illusion because R3 has non-zero contempt play as default which causes it to lose drawn games . Stage 3. OK , but I/RL cannot ponder . Only ponder on games are real games . Stage 4. Even if I/RL can ponder and beats Rybka in ponder on games it's only as strong as Rybka 3 in long time control games . Stage 5 . I don't care if RL beats R3 1t in long time control games . Why are you crippling Rybka by running it on single core only ? Conveniently forgetting that - R3 is sold also as a SP engine so a prospective R3 SP customer needs to be informed that now his money would not be purchasing the strongest SP engine . - RL is as strong as R3 2t . So R3 has advantage over RL only on quads and above which are still a minority hardware . - R3 2t is still not as good as RL because on a dual core RL gives me same quality as R3 2t and leaves one core free on which I can still do other work . I , for one , never run engine matches simultaneously on all cores of any of my machines ( excepting the old resurrected P IV where I have no choice ) . It all reminds me of an old folk tale . The village idiot who could not count properly once had a bet with another villager about how many legs a goat has . The idiot said 3 , the other person said 4 . When the idiot's wife came to know of the bet she said ,"You fool , you've lost . A goat has four legs" . The idiot replied , "Who has lost ? Whatever proof he offers I'll just keep on repeating that a goat has three legs ! " Seems like the technique is still in widespread use . Greetings Azaad, I am struck by how closely your timeline resembles the stages described in Dr. Elisabeth Kubler-Ross' classic groundbreaking book, On Death and Dying. The stages described in that work are as follows: * Denial * Anger * Bargaining * Depression * Acceptance I am sure that you can see the parallels between what you posted and where the various pro-Rybka factions are at particularly as they post in the chess fora. Acceptance is always last. :-) Peace.
Azaad: RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off 1. Silver Opening Suite , 60 out of 100 games played 37.0 - 23.0 in favor of RobboLito ( +18 -4 =38 ) 2. Noomen 2008 opening Suite , 60 out of 60 games played 33.0 - 27.0 in favor of RobboLito ( +15 -9 =36) PGN after the matches are finished .
никел: Good to see you were right! +57 elo
Azaad: Sorry for being so long with the update . RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off 1. Silver Opening Suite , 100 out of 100 games played 55.5 - 44.5 in favor of RobboLito ( +22 -11 =67 ) As you can see Rybka made a great comeback in the last 40 games of the match . Started 2 other 40/40 matches RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off Sedat 600 positions opening suite (*) , 30 out of 600 games played 18.0 - 12.0 in favor of RobboLito ( +9 -3 =18 ) Sedat 200 positions opening suite , 30 out of 400 games played 18.5 - 11.5 in favor of RobboLito ( +10 -3 =17 ) Combined score after 220 games 125 - 95 in favor of RobboLito +48 Elo * I want to settle this argument - whether RobboLito is stronger than R3 at long time controls and if yes then by how much - for ever . A futile hope I know because someone is sure to point out that for long TC match Rybka contempt should be set to default for optimum strength or something like that . But a man can dream .
Azaad: RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off Finished 1. Silver Opening Suite , 100 out of 100 games played 55.5- 44.5 in favor of RobboLito ( +22 -11 =67 ) 2. Noomen 2008 opening Suite , 60 out of 60 games played 33.0 - 27.0 in favor of RobboLito ( +15 -9 =36) Running Sedat 600 positions opening suite , 50 out of 600 games played 30.0- 20.0 in favor of RobboLito ( +15 -5 =30) Sedat 200 positions opening suite , 54 out of 400 games played 34.0 - 20.0 in favor of RobboLito ( +18 -4 =32 ) Combined score after 264 games 152.5 - 111.5 in favor of RobboLito +54 Elo
Azaad: RobboLite 0.085d3 x64 vs Rybka 3 x64 SP ( Contempt Play = 0 ) , 40/40 repeating, Ponder Off Finished 1. Silver Opening Suite , 100 out of 100 games played 55.5- 44.5 in favor of RobboLito ( +22 -11 =67 ) 2. Noomen 2008 opening Suite , 60 out of 60 games played 33.0 - 27.0 in favor of RobboLito ( +15 -9 =36) Running Sedat 600 positions opening suite , 132 out of 600 games played 77.5 - 54.5 in favor of RobboLito ( +33 -10 =89) Sedat 200 positions opening suite , 110 out of 400 games played 61.0 - 49.0 in favor of RobboLito ( +26 -14 =70 ) Combined score after 402 games 227.0 - 175.0 in favor of RobboLito +46 Elo for RobboLito [pre2] Program Elo + - Games Score Av.Op. Draws 1 RobboLite 0.085d3 x64 : 3148 20 20 402 56.5 % 3102 65.2 % 2 Rybka 3 sp : 3102 20 20 402 43.5 % 3148 65.2 %[/pre2] PGN
Gambit_Man: Hi, Thanks for the results Azaad. One of the criticisms that a lot of the CCC crowd have of recent test result is that most people (me included) are testing on a single PC engine vs engine meaning ponder has to be OFF if you want to run the engines with 4 cores each on a quad, or 2 cores each on a dual. They regard ponder OFF testing as not representative of an engine's true perfromance and imply that it in someway disadvantages Rybka. I have posted here a couple of times about a friend of mine who has been trying to test using his normal method of 2 PCs connected - Rybka on one and Ippo/Iggor/Ivanhoe/Firebird on the other with ponder ON. In this set up he has experienced a lot of problems with ppo/Iggor/Ivanhoe/Firebird engines not communicating with the GUI properly He is experiencing continual problems of crashing (resulting in Windows exception errors) and time control losses. He has been using this same testing set up for years now by the way and never any issues before. I have asked whether anyone else is managing to successfully run tests using 2 machines connected and ponder ON but have had no replies ? I have never had these problems using ponder OFF on a single PC so it does seem to be an issue with ponder being ON. I noticed on CCC similar observations have been posted so I guess until those issues can be resolved there will be doubt cast about the significance and credibility of a lot of our testing. I did notice one thing when trying the engines on Arena (I usually use Fritz). The Arena GUI shows the UCI protocol used and what surprised me was that ppo/Iggor/Ivanhoe/Firebird are using the old UCI protocol whereas Rybka, HIARCS and Shredder etc. are using UCI2. In fact even Deep Shredder 9 was using UCI2 so this dates the UCI protocol. I don't know what the specific differences are but maybe the Later Chessbase GUI's are only set up for the UCI2 protocol when running with ponder on ?
Azaad: Gambit_Man Copy/paste from the IPPOLIT thread никел writes: quote: AFAIK RL is not a UCI2 compliant engine, just UCI. What is left to be fully UCI2? UCI_ShowCurrLine UCI_ShowRefutations UCI_LimitStrength UCI_Elo There are others which I'm not sure should be included or not . UCI_EngineAbout -What's the use ? UCI_Chess960 - AFAIK it was introduced with extended UCI protocols . But is it really necessary if an engine is said to be fully UCI2 compliant ? UCI_SetPositionValue - I haven't ever even tried it . Very few engines can be said to be even reasonably UCI2 compliant ( for example I wouldn't call R3 UCI2 compliant ) . In fact I'd use "have UCI2 features" rather than "are UCI2 compliant" . I think "UCI2 compliant" is a word better used with GUI than with the engines .
полная версия страницы