Page 1 of 1
Unbalanced "Elites" Match
Posted: Fri Dec 10, 2010 1:50 am
by kingliveson
Rybka winning the "
Unbalanced Tournament" and IvanHoe coming second, both will now play the same positions used in that tournament with extended time control.
CPU: AMD Phenom II X4 940 @ 3 GHz
Hash: 2048 MB
EGTB: 3-4-5 (256 MB Cache) Nalimov/RobboBases
PawnsHash: 256 MB
Ponder: Off
Time Control: (75m+30s)/40+(15m+30s)
Round 1:
Round 2:
Re: Unbalanced "Elites" Match
Posted: Fri Dec 10, 2010 1:51 am
by kingliveson
Round 3:
Round 4:
Re: Unbalanced "Elites" Match
Posted: Fri Dec 10, 2010 4:10 am
by BB+
Round 1: IvanHoe-R4 0-1:.The dreaded h3 opening. After g4 on move 9, White has weaknesses that are going to require strong piece play to offset. Given the unimprovable king safety issue, White should probably seek to trade major pieces (I don't know if that was really possible). You say "EGTB: 3-4-5 (256 MB Hash) Nalimov/RobboBases", but given that White was showing only a small minus in the 5 piece positions, I would not be so sure that RobboBases were really being used. I don't think this changed the result any.
Round 2: R4-IvanHoe, 1/2. R4 really seems to know how to equalise better in this line. After the trade on c6, the half-open file does nothing for Black, and the structural pawn weakness hurts. IvanHoe did well to close the kingside when Rybka marched the g-pawn up, after which a draw was inevitable.
Round 3: IvanHoe-R4 1-0. The advance Caro-Kann. I don't know how long "book" was followed by the engines on their own (c4 on move 7 looks rare, and maybe 11. Nc3 is the novelty), but White was able to trade off the light-square bishops, and the space advantage derived from e5 was nicely exploited via an eventual kingside push (Black's O-O almost made the king a target, but I can't say that keeping it in the centre looks that great). Black's use of d5 never became important, and the queenside play was essentially zilch. The third incarnation of the Rf6 offer (move 54, after doubling on the f-file) pretty much ended it.
Round 4: R4-IvanHoe 0-1. Castling by White on move 9 (rather than Nh4) looks the more common. Following it up with g4 does not impress me much. This leads to the sort of slow blocked manuevering game that R4 (and Rybka in general) really seems weak at. The bishop pair is not of much value, the g2 bishop is a "big pawn", the d6 pawn is inevitably lost... so after f4 on move 29 Black will have all the play. In contrast to the previous game, White has no play on the kingside, and after much work, Black breaks through on the queenside. The rest of it is "interesting", at least from an old-time view that "computers can't play X", where X might be endgames, or perhaps technical positions. Black eventually makes a passed a-pawn and rolls home. If humans had played it, this might be termed a "strategic crush".
Re: Unbalanced "Elites" Match
Posted: Fri Dec 10, 2010 4:25 am
by kingliveson
BB+ wrote:Round 1: IvanHoe-R4 0-1:.The dreaded h3 opening. After g4 on move 9, White has weaknesses that are going to require strong piece play to offset. Given the unimprovable king safety issue, White should probably seek to trade major pieces (I don't know if that was really possible). You say "EGTB: 3-4-5 (256 MB Hash) Nalimov/RobboBases", but given that White was showing only a small minus in the 5 piece positions, I would not be so sure that RobboBases were really being used. I don't think this changed the result any.
Round 2: R4-IvanHoe, 1/2. R4 really seems to know how to equalise better in this line. After the trade on c6, the half-open file does nothing for Black, and the structural pawn weakness hurts. IvanHoe did well to close the kingside when Rybka marched the g-pawn up, after which a draw was inevitable.
Round 3: IvanHoe-R4 1-0. The advance Caro-Kann. I don't know how long "book" was followed by the engines on their own (c4 on move 7 looks rare, and maybe 11. Nc3 is the novelty), but White was able to trade off the light-square bishops, and the space advantage derived from e5 was nicely exploited via an eventual kingside push (Black's O-O almost made the king a target, but I can't say that keeping it in the centre looks that great). Black's use of d5 never became important, and the queenside play was essentially zilch. The third incarnation of the Rf6 offer (move 54, after doubling on the f-file) pretty much ended it.
Round 4: R4-IvanHoe 0-1. Castling by White on move 9 (rather than Nh4) looks the more common. Following it up with g4 does not impress me much. This leads to the sort of slow blocked manuevering game that R4 (and Rybka in general) really seems weak at. The bishop pair is not of much value, the g2 bishop is a "big pawn", the d6 pawn is inevitably lost... so after f4 on move 29 Black will have all the play. In contrast to the previous game, White has no play on the kingside, and after much work, Black breaks through on the queenside. The rest of it is "interesting", at least from an old-time view that "computers can't play X", where X might be endgames, or perhaps technical positions. Black eventually makes a passed a-pawn and rolls home. If humans had played it, this might be termed a "strategic crush".
Good analysis!! I haven't looked at the games -- just posted the results. As for round 1, yes RobboBases was not being used; it was corrected afterwards. I don't believe either of these engines use internal books, but anything is possible.
Edit: Nevermind, I understand now what you meant by "
I don't know how long 'book' was followed by the engines on their own." By the way, edited Hash to Cache right before your comment went through.
Re: Unbalanced "Elites" Match
Posted: Fri Dec 10, 2010 12:59 pm
by kingliveson
Round 5:
Round 6:
Re: Unbalanced "Elites" Match
Posted: Fri Dec 10, 2010 6:37 pm
by kingliveson
Round 7:
Round 8:
Re: Unbalanced "Elites" Match
Posted: Sat Dec 11, 2010 2:15 am
by kingliveson
Round 9:
Round 10:
Re: Unbalanced "Elites" Match
Posted: Sat Dec 11, 2010 2:24 am
by kingliveson
Final score: Rybka Wins!
Code: Select all
IvanHoe vs Rybka, (75m+30s)/40+(15m+30s)
1 Deep Rybka 4 SSE42 x64 1½00½1½½11 6.0/10
2 IvanHoe 9.49b x64 0½11½0½½00 4.0/10