Houdini 6 vs. Stockfish 6.9.2017

Jouni · Post by **Jouni** » Sat Sep 23, 2017 9:17 am

4 cores 1m+1s HERT openings:

                                    
1   Stockfish 060917 64 BMI2   +22/=68/-10 56.00%   56.0/100
2   Houdini 6 x64-pext         +10/=68/-22 44.00%   44.0/100

+42 ELO for SF. Here's 2 games SF won with both colors.
[pgn]
[Event "championship"]
[Site "?"]
[Date "2017.09.23"]
[Round "5"]
[White "Houdini 6 x64-pext"]
[Black "Stockfish 060917 64 BMI2"]
[Result "0-1"]
[ECO "C34"]
[Annotator "-0.33;-0.65"]
[PlyCount "116"]
[TimeControl "60+1"]

{Intel(R) Core(TM) i5-4460 CPU @ 3.20GHz 3192 MHz W=19.9 plies; 8 393kN/s;
507 151 TBAs B=22.7 plies; 6 989kN/s; 372 226 TBAs} 1. e4 e5 2. f4 exf4 3. Nf3
d6 4. d4 {Both last book move} g5 {-0.65/20 3} 5. Qd3 {-0.33/19 7 (Nc3)} Nc6 {
-0.66/21 5 (Bg7)} 6. Nc3 {-0.33/17 2} Bg7 {-0.71/21 3} 7. d5 {-0.47/20 3} Ne5 {
-0.77/23 2} 8. Nxe5 {-0.59/21 2} Bxe5 {-0.75/21 0} 9. g3 {-0.48/18 2 (Be2)} Qf6
{-0.81/21 4 (a6)} 10. Be2 {-0.78/20 4} Ne7 {-0.80/23 3} 11. Bd2 {-0.74/20 2}
Ng6 {-0.59/23 4 (a6)} 12. O-O-O {-0.57/19 1} a6 {-0.87/22 2} 13. Bh5 {-0.78/21
9 (Kb1)} Bh3 {-0.85/26 3 (Bd7)} 14. Kb1 {-0.71/21 3 (Qf3)} g4 {-0.99/22 1} 15.
Rde1 {-0.76/20 2} O-O-O {-0.80/23 4} 16. Nd1 {-0.63/21 2} Kb8 {-1.10/21 1} 17.
Rhg1 {-0.90/22 6 (Bxg6)} Rde8 {-0.98/25 4} 18. Nf2 {-0.75/21 2} fxg3 {-1.02/25
3} 19. Nxh3 {-0.75/22 2} gxh3 {-0.87/22 0} 20. hxg3 {-0.82/23 3} Bxb2 {-1.01/
25 2} 21. Rh1 {-0.96/23 9} Ne5 {-0.89/23 0 (Rhg8)} 22. Qb3 {-0.91/23 5} Bd4 {
-1.01/24 2} 23. Ref1 {-0.94/20 2} Qe7 {-1.00/23 1 (Qd8)} 24. Qa4 {-0.72/19 1
(Rxh3)} Bb6 {-1.06/26 10 (Bc5)} 25. Rf5 {-0.81/20 3} Rhf8 {-1.08/26 2 (Nd7)}
26. Bh6 {-0.84/21 2} Rg8 {-0.96/25 1} 27. Rxh3 {-1.01/21 3} Ng6 {-1.06/27 1
(Ka7)} 28. Bf3 {-0.88/20 2} f6 {-1.08/23 1} 29. Rh1 {-0.78/20 1 (Bc1)} Ne5 {-1.
04/24 3} 30. Bh5 {-0.92/22 2} Rc8 {-1.04/26 8 (Rd8)} 31. Bf4 {-0.87/20 1} Ng4 {
-0.96/23 0 (Nd7)} 32. Re1 {-0.68/20 3} Rg7 {-1.08/24 2} 33. Qb3 {-0.94/22 6
(Qb4)} Bd4 {-1.19/22 3 (Nf2)} 34. a3 {-0.98/19 3 (Qb4)} Be5 {-1.39/24 5} 35.
Bxg4 {-0.95/19 2} Rxg4 {-1.27/21 0} 36. Bxe5 {-1.18/17 1} fxe5 {-1.37/22 1} 37.
Rf3 {-1.10/18 1} Rf8 {-1.22/24 5} 38. Rc3 {-1.32/19 1} Qd7 {-1.33/23 1} 39. Ka2
{-1.12/19 1} Qb5 {-1.34/26 3 (Rfg8)} 40. Qxb5 {-1.18/21 1} axb5 {-1.44/23 1}
41. Kb3 {-0.97/16 0} Rc8 {-1.43/23 0} 42. Rf3 {-1.09/21 2} h5 {-1.39/25 2 (Ka7)
} 43. Ree3 {-1.12/19 1 (a4)} c6 {-1.45/22 1 (Ka7)} 44. dxc6 {-0.86/18 0 (Rf5)}
bxc6 {-1.49/22 2} 45. Rf7 {-0.80/17 0 (c4)} Rcg8 {-1.76/20 1} 46. Kb4 {-1.75/
20 3 (Rh7)} Rxg3 {-2.55/19 1} 47. Rxg3 {-2.08/20 1} Rxg3 {-2.47/20 0} 48. Ka5 {
-2.44/20 1 (Rd7)} Kc8 {-2.70/21 1} 49. Kb6 {-2.28/19 0} Rc3 {-2.92/20 0} 50.
Rc7+ {-2.95/19 1} Kd8 {-3.35/22 1} 51. Rh7 {-3.18/20 1} Rxc2 {-3.65/20 0} 52.
Rxh5 {-3.76/20 2} Kd7 {-4.02/22 1} 53. Rh7+ {-3.88/19 1} Ke6 {-4.08/19 0} 54.
Rh6+ {-3.89/19 1} Ke7 {-4.21/21 1} 55. Rh3 {-4.39/20 1} Rc4 {-4.75/21 1} 56.
Rb3 {-4.53/19 1 (Rh7+)} Ke6 {-5.70/22 1} 57. Rd3 {-4.97/21 1 (Rh3)} b4 {-6.77/
23 1 (Rd4)} 58. a4 {-6.31/21 1} b3 {-6.96/22 0} 0-1

[Event "championship"]
[Site "?"]
[Date "2017.09.23"]
[Round "6"]
[White "Stockfish 060917 64 BMI2"]
[Black "Houdini 6 x64-pext"]
[Result "1-0"]
[ECO "C34"]
[Annotator "-0.48;-0.39"]
[PlyCount "87"]
[TimeControl "60+1"]

{Intel(R) Core(TM) i5-4460 CPU @ 3.20GHz 3192 MHz W=21.0 plies; 7 352kN/s;
117 818 TBAs B=19.0 plies; 8 283kN/s; 92 439 TBAs} 1. e4 e5 2. f4 exf4 3. Nf3
d6 4. d4 {Both last book move} g5 {-0.39/18 3} 5. h4 {-0.48/20 5 (Nc3)} g4 {-0.
46/18 1} 6. Ng1 {-0.42/21 3} f5 {-0.49/18 3} 7. Ne2 {-0.21/20 2 (Qd3)} fxe4 {
-0.14/19 4} 8. Nxf4 {-0.10/21 4} Nf6 {-0.08/20 6} 9. d5 {0.00/21 2} Qe7 {-0.21/
19 2} 10. Nc3 {-0.15/19 0 (Bc4)} Bg7 {0.37/19 4 (a6)} 11. Bb5+ {0.20/18 3} Kd8
{0.32/19 2 (c6)} 12. Be3 {0.62/19 2 (0-0)} a6 {0.73/20 6 (g3)} 13. Bc4 {1.12/
20 3} Re8 {0.95/22 14 (g3)} 14. Rf1 {1.46/20 2 (0-0)} b5 {1.35/20 5 (Qf7)} 15.
Bb3 {1.68/20 2} Bb7 {1.48/21 4} 16. Ne6+ {2.10/23 8} Kc8 {1.56/21 3} 17. a4 {
1.90/25 6 (Nxg7)} b4 {1.23/17 1} 18. Ne2 {1.93/22 1} Bh8 {1.42/19 4 (a5)} 19.
a5 {2.67/19 1 (Qd2)} c6 {1.45/19 5 (h6)} 20. dxc6 {2.80/21 1 (N2d4)} Nxc6 {2.
05/20 4} 21. Bf4 {3.05/22 2} d5 {2.16/20 2 (Ne5)} 22. Nc7 {3.28/22 5} Ne5 {2.
38/19 1} 23. Nxe8 {3.27/20 0} Nxe8 {2.37/18 0} 24. Bg5 {3.52/22 5} Nf6 {2.50/
21 2 (Qc5)} 25. Qd2 {3.74/22 2 (Bxd5)} Nf3+ {2.75/19 2} 26. gxf3 {3.87/22 1}
exf3 {3.05/21 3} 27. Ra4 {3.91/23 2} h6 {2.73/20 1} 28. Qe3 {3.96/22 0} Qxe3 {
2.82/20 2} 29. Bxe3 {4.12/23 1} Kd7 {2.64/20 2 (fxe2)} 30. Rxb4 {4.05/22 3} Re8
{2.94/19 2} 31. Rxb7+ {4.04/19 1} Kc8 {2.93/18 0} 32. Ng3 {4.29/22 3} Kxb7 {3.
39/19 3} 33. Kf2 {4.47/21 2} h5 {3.68/19 2} 34. Rd1 {4.82/22 2} Re7 {3.75/17 1
(Re5)} 35. Bxd5+ {5.65/21 2 (c3)} Nxd5 {4.06/18 1} 36. Rxd5 {5.30/21 0} Bxb2 {
4.10/19 1} 37. Rxh5 {5.90/21 1} Be5 {4.32/19 1 (Rc7)} 38. Bc5 {6.48/20 1} Re6 {
4.54/19 1} 39. Rg5 {6.50/20 0} Bc7 {4.60/18 1} 40. Rxg4 {6.79/22 1} Bxa5 {4.64/
17 1} 41. h5 {7.04/20 1} Bd2 {4.69/17 1 (Rc6)} 42. Kxf3 {7.56/21 3} Rf6+ {5.12/
16 2 (a5)} 43. Ke4 {7.85/19 1} Rf7 {5.48/18 1 (Rc6)} 44. Bd4 {8.10/20 1 (Rg6)}
1-0

[/pgn]

Jouni · Post by **Jouni** » Sat Sep 23, 2017 4:09 pm

Same test with one core ended:

Code: Select all



                                    
1   Houdini 6 x64-pext         +22/=65/-13 54.50%   54.5/100
2   Stockfish 060917 64 BMI2   +13/=65/-22 45.50%   45.5/100

So +31 for Houdini. Obviously more games needed as always

.

Laskos · Post by **Laskos** » Sat Sep 23, 2017 4:19 pm

Jouni wrote:Same test with one core ended:
Code: Select all
                                    
1   Houdini 6 x64-pext         +22/=65/-13 54.50%   54.5/100
2   Stockfish 060917 64 BMI2   +13/=65/-22 45.50%   45.5/100
So +31 for Houdini. Obviously more games neede as always .

No, these games are enough. Your first match is conclusive (t>2), but the result is weird. Weird and conclusive is the difference between the two matches. I don't know what to say, at the first glance H6 scales badly with SMP, but I doubt it.

Nordlandia · Post by **Nordlandia** » Sat Sep 23, 2017 4:51 pm

Kai Laskos: do you believe negative contempt is appropriate for K11.2.2 if facing H6?

Contempt range between -2 to -3 might yield more draws.

Considering H6 is somewhat stronger.

Laskos · Post by **Laskos** » Sat Sep 23, 2017 5:09 pm

Nordlandia wrote:Kai Laskos: do you believe negative contempt is appropriate for K11.2.2 if facing H6?

Contempt range between -2 to -3 might yield more draws.

Considering H6 is somewhat stronger.

Yes. The issue with negative Contempt is that in say a head-to-head match (like TCEC superfinal), one can diminish the magnitude of the loss in a match, but one cannot win using negative contempt when he is anyway losing. So, even if your engine is objectively weaker, in simple head-to head match for the final win, if the magnitude of possible loss doesn't interest you, you better leave Contempt=0. You will have more chances to win the match, being weaker, because of more statistical flukes.

OTOH for rating lists, one has to precisely calibrate negative or positive Contempt to have the best rating, there the magnitude of loss and win is important.

Nordlandia · Post by **Nordlandia** » Sat Sep 23, 2017 5:26 pm

Laskos wrote:
Nordlandia wrote:Kai Laskos: do you believe negative contempt is appropriate for K11.2.2 if facing H6?

Contempt range between -2 to -3 might yield more draws.

Considering H6 is somewhat stronger.
Yes. The issue with negative Contempt is that in say a head-to-head match (like TCEC superfinal), one can diminish the magnitude of the loss in a match, but one cannot win using negative contempt when he is anyway losing. So, even if your engine is objectively weaker, in simple head-to head match for the final win, if the magnitude of possible loss doesn't interest you, you better leave Contempt=0. You will have more chances to win the match, being weaker, because of more statistical flukes.

OTOH for rating lists, one has to precisely calibrate negative or positive Contempt to have the best rating, there the magnitude of loss and win is important.

Code:
Houdini 6 - Komodo 11.2.2, Blitz 3m+2s 2017

1 Houdini 6 Pro x64-popc +63 +16/=27/-7 59.00% 29.5/50
2 Komodo 11.2.2 64-bit -63 +7/=27/-16 41.00% 20.5/50

<----------------->

Komodo 11.2.2 may achieve more draws in head-to-head match with negative contempt.

This is my point. Final score might become more narrowed with more draw than otherwise lost games.

fauzi · Post by **fauzi** » Sun Sep 24, 2017 1:54 pm

Can you please share the pgn?

Milos · Post by **Milos** » Tue Sep 26, 2017 12:15 am

Laskos wrote:
Jouni wrote:Same test with one core ended:
Code: Select all
                                    
1   Houdini 6 x64-pext         +22/=65/-13 54.50%   54.5/100
2   Stockfish 060917 64 BMI2   +13/=65/-22 45.50%   45.5/100
So +31 for Houdini. Obviously more games neede as always .
No, these games are enough. Your first match is conclusive (t>2), but the result is weird. Weird and conclusive is the difference between the two matches. I don't know what to say, at the first glance H6 scales badly with SMP, but I doubt it.

Since the difference is 73Elo total it seams as if H6 was not using 4 cores but only 2.

Houdini 6 vs. Stockfish 6.9.2017

Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017

Re: Houdini 6 vs. Stockfish 6.9.2017