|
< Earlier Kibitzing · PAGE 10 OF 79 ·
Later Kibitzing> |
| May-01-06 | | Dionyseus: Rybka 1.2 coming out tomorrow on May 1! http://www.rybkachess.com/index.php... |
|
| May-01-06 | | notyetagm: Rybka eval, please. The following position is from the incredible ending of Naiditsch vs Stellwagen, 2006, from this year's Corus B tournament. White to play.
click for larger viewDoes Rybka find the fantastic winning line that Naiditsch actually played, 35 ♖xc6!! ♔xc6 36 ♗xh6!!. The point of White's sacrifice is that he has made advanced connected passed g- and h-pawns that are well-supported by the White h4-king and opposed only by the Black rook, which is powerless to stop them by itself and gets no help from the offside Black c6-king. After 36 ... ♖xh6 37 g5 White queens a pawn.
click for larger viewThe best example I have ever seen of the well known endgame fact that <advanced connected passed pawns beat a rook> and a great example of <do not look at what comes off the board, look at what is left on the board>. |
|
| May-01-06 | | Dionyseus: <NotYetAGM> After 18 minutes Rybka 1.1 felt that Bxh6 is the strongest move and evaluates it as 1.12 at a depth of 22. At 26 minutes it still felt that Bxh6 was the strongest move, at a depth of 23. After I showed it Naiditsch's line, it quickly discovered how strong it is. I'm now checking to see if Shredder 9.1 can find the move by its own. After 1 minute it considers Bxh6 the strongest move evaluating it 1.37 at depth 19. I'm going to let it calculate for 30 minutes. |
|
| May-01-06 | | Dionyseus: After 30 minutes Shredder 9.1 felt that Bxh6 was strongest, evaluates it at 3.07 at depth 27. Apparently 35.Bxh6 wins too. 35.Bxh6 Nxe5 36.Bf4 Ng6+ 37.Kg5 Nxf4 38.Kxf4 Rf7+ 39.Ke5 Rg7 40.Rg1 and it's clear that black can't stop white's advanced pawns. Naiditsch's line still looks nicer though ;p |
|
| May-01-06 | | notyetagm: <Dionyseus> Thanks. I wonder if any chess engine can find Naiditsch's incredible winning line. Hydra, anyone?
:-)
|
|
| May-01-06 | | Dionyseus: <notyetagm> Rybka 1.2 came out this afternoon and I decided to check out whether this new version can find Naiditsch's sacrifice, and it turns out that it does! After 18 minutes it says Rxc6 is the strongest move, and evaluates it at +0.86 at a depth of 25 after calculating through about 228 million positions. At 19 minutes it increases its evaluation slightly to +0.89 at a depth of 26. At 1 hour and 13 minutes it no longer thinks Rxc6 is strongest, preferring Bxh6 and evaluating it at +1.74 at a depth of 27 after calculating through about 857 million positions. The fact that it found Rxc6 as a strong move is definitely an improvement:) |
|
| May-01-06 | | Hesam7: <Dionyseus> Can I ask what your hardware is?? |
|
| May-01-06 | | Dionyseus: <Hesam7> Pentium 4 3.2ghz with hyperthreading (although Rybka does not yet support multiple processors/threading), with 1gb of ram. My next computer will be an AMD machine, preferably with a dual-core cpu, AMDs perform better than Pentiums with chess engines. |
|
| May-01-06 | | Hesam7: <Rybka does not yet support multiple processors/threading> This is a very bad weakness. Fruit does not support it either. How many nodes/sec do you get with this hardware?? You might try the Pentium Extreme Edition. It has two cores each clocked at 3.73 GHz and hyperthreading is available on both cores so your system sees 4 logical processors. |
|
| May-01-06 | | Dionyseus: <Hesam7> It's a big weakness indeed, which makes Rybka's success even more remarkable. If it's so good even though it is currently limited to a single processor and thread, imagine how much more powerful it will be once it supports multi processors and threads.
As for how many nodes Rybka goes through per second, it depends on the position, but 1.2 is definitely faster now than 1.1. For that Naiditsch rook sacrifice test position, it averaged 205k nodes. In most positions it averages less than 100k nodes. |
|
| May-01-06 | | Hesam7: <Dionyseus: As for how many nodes Rybka goes through per second, it depends on the position, but 1.2 is definitely faster now than 1.1. For that Naiditsch rook sacrifice test position, it averaged 205k nodes. In most positions it averages less than 100k nodes.> Wow then it should do a lot of computation per node. The evaluation function should be pretty complex. Which explains it unprecedented success. |
|
| May-01-06 | | Hesam7: From www.Rybkachess.com
<Q. What is next for Rybka?
A. Please stay tuned for an upcoming announcement in early-to-mid May about future Rybka versions. Prospective Rybka customers should be aware that Rybka 1 is coming to the end of its cycle, and that Rybka 2 _may_ go on sale as early as June 1, 2006. Those who would like to plan their software purchases are advised to wait until this announcement so that they can make informed decisions. Q. When will we see a multi-processor version of Rybka? A. This is right around the corner - please wait for the announcement mentioned above.> |
|
| May-02-06 | | notyetagm: <Dionyseus: <notyetagm> Rybka 1.2 came out this afternoon and I decided to check out whether this new version can find Naiditsch's sacrifice, and it turns out that it does!> So Rybka 1.2 does find Naiditsch's brilliant idea. Too bad it does not give it a stronger eval and later switches to the weaker 35 ♗xh6. Do you care to run this test on other chess engines that you may have? Shredder is supposed to be good at endgames; this position might be ideal for it. |
|
| May-02-06 | | jhoro: I ran Spike, Ktulu and few other engines using Arena. Most think that 35.Bxh6 is best. These are some of the lines at depth 18/19: Ktulu 4.2:
18 04:01 213.704.615 173.703 +1.50 Bf4xh6 Nc6xe5 Bh6f4 Ne5f3+ Kh4g3 e6e5 Kg3xf3 Rh7f7 h5h6 Rf7xf4+ Kf3g3 Rf4f8 Rc1e1 Rf8h8 Re1xe5 Rh8xh6 Re5xd5+ Rh6d6 Rd5xd6+ Kd7xd6 g4g5 Kd6e5 g5g6 19 05:37 296.834.753 116.118 +1.39 Bf4xh6 Nc6xe5 Bh6f4 Ne5f3+ Kh4g3 e6e5 Kg3xf3 Rh7f7 h5h6 Rf7xf4+ Kf3g3 Rf4f8 g4g5 Rf8h8 Kg3f3 e5e4+ d3xe4 Rh8f8+ Kf3e2 d5xe4 Rc1c4 d4d3+ Ke2e3 Rf8e8 Spike1.1:
18+ 03:43 165.650.937 746.541 +0.27 Bf4xh6 Rh7xh6 g4g5 Rh6h7 Rc1xc6 Kd7xc6 g5g6 Rh7g7 Kh4g5 18 06:08 280.361.494 764.090 +1.10 Bf4xh6 Nc6xe5 Bh6f4 Ne5g6+ Kh4g5 Ng6xf4 Kg5xf4 Kd7d6 Kf4g5 Rh7g7+ Kg5h4 e6e5 g4g5 e5e4 h5h6 Rg7e7 g5g6 e4xd3 h6h7 d3d2 Rc1d1 Re7e4+ Kh4g5 Re4e5+ Kg5g4 Re5e4+ Kg4f5 Spike1.0A came the fastest with Bxh6 in about 2.5 min on a 2.6GHz P4. |
|
| May-02-06 | | jhoro: Naiditsch line may not be the best. Simply Stellwagen did not defend well enough. In particular 36...Rxh6. Instead 36...Kd7 seems to be better according to the engines. Some lines after 35.Rxc6 Kxc6 36.Bxh6 ... Ktulu:
< 17 03:35 196.708.849 115.560 -1.26 Kc6d7 Bh6g5 Kd7e8 Bg5f6 Ke8f8 g4g5 Rh7c7 g5g6 Rc7c1 h5h6 Rc1h1+ Kh4g5 Rh1g1+ Kg5f4 Rg1f1+ Kf4g4 Rf1g1+ Kg4f3 Rg1xg6 h6h7 Rg6h6 h7h8Q+ Rh6xh8 18 04:21 237.578.711 87.271 -1.06 Kc6d7 Bh6g5 a7a6 Bg5f6 b6b5 a4xb5 a6xb5 g4g5 b5b4 g5g6 Rh7e7 g6g7 Re7xg7 Bf6xg7 b4b3 h5h6 b3b2 h6h7 b2b1Q h7h8Q Qb1xd3> Spike:
< 18 00:51 39.155.859 765.645 -0.90 Kc6d7 Bh6g5 a7a5 Bg5d2 b6b5 a4xb5 a5a4 g4g5 a4a3 g5g6 Rh7g7 Bd2c1 a3a2 Bc1b2 Rg7g8 Kh4g5 Kd7e7 Bb2xd4 Rg8c8 19 01:09 52.828.420 759.946 -1.06 Kc6d7 Bh6g5 a7a5 Bg5d2 Kd7e8 g4g5 Rh7f7 g5g6 Rf7f1 h5h6 Ke8f8 Bd2g5 Rf1h1+ Kh4g3 Rh1g1+ Kg3f4 Rg1h1 g6g7+ Kf8g8 Kf4f3 Rh1a1 Bg5d8 Ra1xa4 Bd8xb6 19 01:42 77.107.901 755.145 -1.06 Kc6d7 Bh6g5 a7a5 Bg5d2 Kd7e8 g4g5 Rh7f7 g5g6 Rf7f1 h5h6 Ke8f8 Bd2g5 Rf1h1+ Kh4g3 Rh1g1+ Kg3f4 Rg1h1 g6g7+ Kf8g8 Kf4f3 Rh1a1 Bg5d8 Ra1xa4 Bd8xb6 19 06:59 377.581.581 81.111 -1.32 Kc6d7 Bh6g5 a7a6 Bg5f6 b6b5 a4xb5 a6xb5 g4g5 b5b4 g5g6 Rh7e7 g6g7 Re7xg7 Bf6xg7 b4b3 h5h6 b3b2 h6h7 b2b1Q h7h8Q Qb1xd3 Qh8f8
>
|
|
| May-02-06 | | RandomVisitor: <jhoro>Rybka thinks that 36...Kd7 37.Bd2 is eventually fatal for Black. |
|
| May-02-06 | | Dionyseus: <notyetagm> Well Bxh6 is not necessarily weaker than Rxc6, both lines clearly win. Naiditsch's line simply looks a lot nicer to us humans ;p |
|
May-02-06
 | | offramp: I have Rybka but I can't get it to work with Chessbase - any ideas? |
|
| May-02-06 | | Dionyseus: <offramp> Go to folder Programs Files --> Chessbase --> Engines --> UCI --> make folder Rybka, and place the exe file in it. Now on your Chessbase program, go to Engine menu, go to Create UCI Engine, and then browse to the Rybka folder where you put the exe file at and you're all set. |
|
| May-02-06 | | notyetagm: <Dionyseus: <notyetagm> Well Bxh6 is not necessarily weaker than Rxc6, both lines clearly win. Naiditsch's line simply looks a lot nicer to us humans ;p> Yes, that thought has dawned on me. The comps like ♗xh6! because it is more "efficient" while us humans like ♖xc6! because it is "flashier". |
|
| May-02-06 | | Dionyseus: My blitz engine tournament with the new Rybka 1.2 just finished, here are the results: Hardware:
Pentium 4 3.2ghz with Hyperthreading
1gb ram
200mb ram available to each engine
No Permanent Brain
First round, 1 cycle round robin, blitz 5':
1. Rybka 1.2 5.5/7
2. Rybka 1.0 Beta 5.0/7
3. Rybka 1.1 5.0/7
4. Shredder 9.1 3.5/7
5. Fritz 9 3.5/7
6. Deep Shredder 9.1 2.5/7
7. X3D Fritz 2.5/7
8. Fruit 2.2.1 0.5/7
4th place tiebreak blitz 5':
Shredder 9.1 2.0/4
Fritz 9 2.0/4
4th place tiebreak 2 blitz 5:
Shredder 9.1 1.5/2
Fritz 9 0.5/2
Semifinals 1 6 game match blitz 5':
Rybka 1.0 Beta 4.0/6
Shredder 9.1 2.0/6
Semefinals 2 6 game match blitz 5':
Rybka 1.2 3.5/6
Rybka 1.1 2.5/6
Third Place 6 game match blitz 5':
Rybka 1.1 4.5/6
Shredder 9.1 1.5/6
Finals 8 game match blitz 5':
Rybka 1.2 5.0/8
Rybka 1.0 beta 3.0/8
notes: I noticed that Deep Shredder 9.1 didn't perform so well in my blitz tournament because whenever it was its turn to move its first second or so of calculation is really slow (and by slow I mean less than 10k N per second) as it starts to utilize both processor threads. conclusion: Rybka 1.2 is a mighty powerful blitz engine. I'll try a slower time control tournament soon. |
|
| May-02-06 | | whatthefat: <Dionyseus>
That's a pretty convincing victory for Rybka 1.2. I look forward to seeing some results at a slower time control. I also fear what Rybka 2 will be capable of! |
|
| May-02-06 | | THE pawn: Can someone explain to me why my version of Rybka (1.00 beta version 32 bits) is awfully bad?? I'm running it on Arena, so that may explain his incapacity to analyze correctly, unless I'm mistaken. Three games out of four he lost against fruit 2.1 and he lost all games I made him play against chessmaster 9k, which is not normal we all agree. Also, he does have any opening repertoire. Can somebody enlighten me on this?? thx in advance. |
|
| May-02-06 | | sharpnova: Bxh6 is a bit faster than Rxc6..
<notyetagm> is clearly of lower chess calibre than the rest of us. |
|
| May-02-06 | | Dionyseus: <The Pawn> Get a Chessbase interface. The opening book that comes with the beta is called The Turk.ctg, you need to put that file in the folder Books in your chessbase folder. |
|
 |
|
|
< Earlier Kibitzing · PAGE 10 OF 79 ·
Later Kibitzing> |
Kibitzing
