Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Deep Fritz 12 mini review
- - By Carl Bicknell (*****) Date 2010-03-01 18:34
I meant to do this several days ago and in more depth, but work has been busy.
What follows is a review based mainly on the engine DF12 on matters not covered by CEGT etc e.g. playing style, distinctive points about the program etc.

Playing Style
Style of play seems to be a little more definite than Fritz 11, but not as aggressive as Deep Fritz 10.1.
Here's a nice example of one of DF12 more aggressive games:

[White "Deep Fritz 12"]
[Black "Hiarcs 10"]
[Result "1-0"]

{Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz 3400 MHz  W=18.1 plies;  B=12.3 plies.
1. d4 {B 0} e6 {B 0} 2. e4 {B 0} d5 {B 0} 3. Nc3 {B 0} Nf6 {B 0} 4. e5 {B 0}
Nfd7 {B 0} 5. f4 {B 0} c5 {B 0} 6. Nf3 {B 0} Nc6 {B 0} 7. Be3 {B 0} cxd4 {B 0}
8. Nxd4 {B 0} Bc5 {B 0} 9. Qd2 {B 0} O-O {B 0} 10. O-O-O {B 0} a6 {B 0} 11. h4
{B 0} Nxd4 {B 0} 12. Bxd4 {B 0} b5 {B 0} 13. Rh3 {B 0} b4 {B 0} 14. Na4 {B 0}
Bxd4 {B 0} 15. Qxd4 {B 0} a5 {B 0} 16. Bb5 {B 0} Rb8 {B 0} 17. Bd3 {B 0} Bb7 {
B 0} 18. h5 {0.19/18 13} f6 {-0.15/12 14 (h6)} 19. Rg3 {0.57/19 15 (exf6)} Rf7
{-0.06/12 17 (h6)} 20. Bg6 {0.87/20 8 (h6)} hxg6 {0.25/15 75 (Re7)} 21. hxg6 {
3.33/17 14} Rc8 {2.01/11 5} 22. Rh1 {3.43/16 1 (gxf7+)} Nf8 {2.01/11 6} 23.
gxf7+ {3.33/16 1} Kxf7 {1.94/11 2} 24. Kb1 {3.34/18 12 (Rh4)} Qc7 {2.39/13 18}
25. c3 {3.55/18 5} Qc4 {2.41/12 5} 26. exf6 {3.63/18 6} gxf6 {2.37/12 5 (Qxd4)}
27. Qd1 {3.77/19 7 (Nb6)} Rc7 {2.10/11 9 (Qe4+)} 28. Qh5+ {3.76/17 5} Ke7 {2.
35/11 1} 29. Rg7+ {4.01/18 8} Kd6 {2.60/12 7} 30. Rxc7 {3.99/19 10} Kxc7 {3.05/
13 19 (Qxc7)} 31. cxb4 {4.32/19 7} Qe4+ {2.93/12 4} 32. Ka1 {4.44/20 7} Qxb4 {
3.23/12 7 (Nd7)} 33. Rc1+ {4.59/19 2} Kd8 {3.78/13 6 (Kd6)} 34. Nc5 {5.30/19 3}
Ke7 {4.52/13 16} 35. Qh6 {7.11/18 4} Kf7 {5.31/13 8} 36. g4 {8.12/18 2} Bc8 {
5.86/13 11 (a4)} 37. Nd3 {10.55/18 1} Qb7 {7.78/13 10} 38. g5 {11.22/18 1} fxg5
{9.34/13 12} 39. Ne5+ {11.15/17 1} Kg8 {9.43/13 2} 40. Qxg5+ {11.88/18 1}

1-0

However, note that the above was mostly due to the book: move 18 was the last book move.

I have also noted a tendency to have a Fritz-like obsession over central control which gives some nice wins (especially over Hiarcs for some reason).

Other notable points
The program plays most endgames exceptionally well. Shredder is known for this, but Fritz is also very good in certain positions.

The program must be pruning very heavily because it searches VERY deep, very fast. It's common on my machine seeing depths in the high 20's even in the middlegame. Truely the name Deep Fritz 12 is justified! Assuming the depths shown can be compared to other versions of Fritz there has certainly been progress made here.

Tactical Ability
Now we get on to my favourite subject: tactics. I have been building test suites for 15 years or so and as hardware has improved many of my favourite positions have had to be shelved because they were no longer tough enough. However, I now have about 150 positions which can still give the best progs a workout.
And now the shock. If the last version of Fritz you owned was DF10.1 then you're not going to like this: Deep Fritz 12 is a sissy when it comes to tactics. I know the word Fritz was once interchangeable with 'Tactical Monster' etc but no longer. DF12 even managed to knock Shredder off the bottom post (also a notoriously weak tactician - though very strong program - as I indicated on a review of it some months ago).

Let me give an example:

 
rnbq1rk1/ppp2ppp/4pn2/8/1b1PP3/2NB1N2/PP3PPP/R1BQK2R w KQ - 0 0


This position comes from one of my games and is probably the easiest position in the testsuite. Most programs will get this in a couple of seconds. So I was rather horrified to see DF12 on a quad Core i7 @ 3.4 GHz taking over 40 seconds! You can imagine how it struggled with harder positions...

Many have commented that tactical strength and elo are separate issues. Well they are to an extent. Sure DF12 is much stronger than DF10.1 (which is strong tactically, even by today's standards). But I also observe games where Rybka 3 gets a good position against DF12 and kills it with a haymaker. And when DF12 gets a good position against R3...R3 finds a shot to get perpetual. So they are not ENTIRELY separate issues. Also, for human analysis tactics are vital - that's often why we switch the computer on! For example:

r2q2kr/5np1/2p2p2/2Bb4/P4P2/2Q3R1/2P3PP/R4BK1 w - - 0 34


This is from one of my games. White is winning here but I couldn't see a way to end things so I played 34 a5 and eventually won. But Rybka finds 34.Rg7! Kg7 35.Qg3 Kh6 36. Qh3 Kg7 37. Qg4 Kh6 38. Ra3! winning! What a neat way to end the game! Wish I'd seen that! DF12 doesn't get it...

Anyway here are some scores from the testsuite:
(I've tested about 20 progs but here is a selection)

Rybka 3 = 90%
Bright 0.4a = 86%
Zappa Mexico II = 82%
Deep Fritz 10.1 = 80%
HIARCS 12.1 = 80%
Deep Junior 2010 = 72%
Deep Fritz 12 = 64%

The only new program I haven't tested is Spark, which I hear is excellent.
Anyway, disappointing from Fritz 12.

The opening book isn't bad at all, although not as ambitious as either Rybka 3's or Hiarcs 12's. The GUI is VERY ambitious, basically copying Microsoft Office 2007 in terms of style (and that's a big compliment, it looks really good) but there are several bugs in it which detract from the overall experience.

The main problem with DF12 is the price. Yes it has loads in it. But so does Fritz 12 at half the price. You're being charged double for an engine which is about 30 elo stronger than DF11 and obviously weaker than Stockfish. Roll on the days when programmers only charge one price and all engines are multi core...
Parent - - By Uly (Gold) Date 2010-03-01 21:18

> basically copying Microsoft Office 2007


Aquarium copied it first, then Fritz copied Aquarium :)

Thanks for the review! I skipped DF11 and I think I'll skip DF12 as well, do you have planned to keep doing reviews like this for other engines?
Parent - - By Carl Bicknell (*****) Date 2010-03-01 21:27 Edited 2010-03-01 21:40

> do you have planned to keep doing reviews like this for other engines?


Yes, when I have time! Which one did you have in mind?

UPDATE: Spark 0.3a gets 84% tactics. The ones it solves are solved very fast, typically faster than Rybka 3. Unfortunately it has some 'blind spots' esp in endgame problems which lowered its score.

Overall I'd say R3 maybe just has a slight edge.
Parent - - By Uly (Gold) Date 2010-03-01 21:28
As you said on your review, Spark, I'd like to know if the tactical hype about it is true.
Parent - - By Master Om (Bronze) Date 2010-03-02 02:51 Edited 2010-03-02 02:54
New game
rnbq1rk1/ppp2ppp/4pn2/8/1b1PP3/2NB1N2/PP3PPP/R1BQK2R w KQ - 0 1


Analysis by spark-0.3a-vc64-mp:

1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qc1
  +-  (1.51)   Depth: 6/6   00:00:00  3kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2
  +-  (1.55)   Depth: 7/6   00:00:00  6kN
1.e5 Bxc3+ 2.bxc3 Nd5 3.Bxh7+ Kxh7 4.Ng5+ Kg6 5.h4 Qe7 6.Qd3+ f5 7.h5+ Kh6 8.Nxe6+ Kh7 9.Nxf8+ Qxf8
  +/-  (1.22)   Depth: 8/12   00:00:00  33kN
1.e5
  +-  (1.47)   Depth: 9/12   00:00:00  47kN
1.e5 Nd5 2.Qc2 h6 3.a3 Ba5 4.0-0 Nc6 5.Be3
  +/=  (0.67)   Depth: 9/14   00:00:00  168kN
1.0-0 c5 2.e5 cxd4 3.Nxd4 Bc5 4.Nb3 Nfd7 5.Bf4
  +/-  (1.05)   Depth: 9/14   00:00:00  200kN
1.0-0 h6 2.Be3 Nc6 3.a3 Be7 4.Qb3 a6 5.Rfc1 Qd6
  +/-  (1.02)   Depth: 10/14   00:00:00  338kN
1.0-0 h6 2.a3 Be7 3.Bf4 c5 4.dxc5 Bxc5 5.b4 Bb6 6.Ne5 Nc6
  +/-  (1.10)   Depth: 11/14   00:00:00  517kN
1.e5
  +/-  (1.28)   Depth: 11/24   00:00:00  561kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.a3 Be7 6.0-0 Nc6
  +-  (1.62)   Depth: 11/24   00:00:00  598kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.a3 Be7 6.Qf4 f5 7.exf6+ Rxf6 8.Qg4+ Kh8
  +-  (1.56)   Depth: 12/24   00:00:00  659kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.a3 Be7 6.Qf4 Bg5 7.Nxg5 hxg5
  +-  (1.62)   Depth: 13/26   00:00:00  783kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.Qe3 Rg8 6.0-0 Nc6 7.a3 Be7 8.Qf4
  +-  (1.64)   Depth: 14/28   00:00:00  1314kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.Qe3 Rg8 6.0-0 Nc6 7.a3 Be7 8.Qe4 Kf8 9.Qxh7 Nxd4 10.Qxh6+ Rg7
  +-  (1.59)   Depth: 15/30   00:00:00  1630kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.Qe3 Nd7 6.0-0 c6 7.Qf4 f5 8.exf6+ Nxf6 9.Qg3+ Kh8 10.Rfe1
  +-  (1.64)   Depth: 16/30   00:00:00  2329kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.Qe3 Nd7 6.0-0 c6 7.Bc2 f5 8.exf6+ Qxf6 9.Rae1 Nb6 10.Ne5
  +-  (1.61)   Depth: 17/36   00:00:00  3512kN
1.e5 Ng4 2.h3 Nh6 3.Bxh6 gxh6 4.Qd2 Kg7 5.Qf4 Nd7 6.0-0 f5 7.exf6+ Nxf6 8.Rac1 c6 9.Qg3+ Kh8 10.Rfe1 Nd5 11.Ne5
  +-  (1.55)   Depth: 18/38   00:00:01  8153kN

(, Prakash 02.03.2010)
A staggering 2 seconds to find the move!!

New game
r2q2kr/5np1/2p2p2/2Bb4/P4P2/2Q3R1/2P3PP/R4BK1 w - - 0 1


Analysis by spark-0.3a-vc64-mp:

1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6
  +-  (3.28)   Depth: 6/9   00:00:00  9kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6
  +-  (3.28)   Depth: 7/10   00:00:00  19kN
1.a5
  +-  (3.56)   Depth: 8/14   00:00:00  36kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Nb5
  +-  (3.56)   Depth: 8/14   00:00:00  39kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Nb5 5.Qe1
  +-  (3.52)   Depth: 9/16   00:00:00  62kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Nb5 5.Qe1 Qd7 6.h3
  +-  (3.58)   Depth: 10/17   00:00:00  105kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Nb5 5.Qe1 Qd7 6.h3
  +-  (3.58)   Depth: 11/19   00:00:00  159kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Nb5 5.Qe1 Qd7 6.h3 g5 7.Qa1
  +-  (3.60)   Depth: 12/19   00:00:00  258kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Nb5 5.Qe1 Qd7 6.h3 g5 7.Qa1
  +-  (3.60)   Depth: 13/23   00:00:00  482kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Qc7 5.a7 Nb5 6.Qa5 Qxa5 7.Rxa5 Nc3
  +-  (3.57)   Depth: 14/23   00:00:00  1675kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Qc7 5.a7 Nb5 6.Qa5 Qxa5 7.Rxa5 Rxf4 8.Bxb5 cxb5 9.Rxb5
  +-  (3.62)   Depth: 15/31   00:00:00  5856kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Qc7 5.a7 Nb5 6.Qa5 Qxa5 7.Rxa5 Rxf4 8.Bxb5 cxb5 9.Rxb5
  +-  (3.62)   Depth: 16/36   00:00:02  15773kN
1.a5 Nd6 2.Bd3 Rh4 3.Ra4 Kh8 4.a6 Qc7 5.a7 Nb5 6.Qa5 Qxa5 7.Rxa5 Rxf4 8.Bxb5 cxb5 9.Rxb5 Kh7
  +-  (3.67)   Depth: 17/40   00:00:05  39296kN
1.Rxg7+ Kxg7 2.Qg3+ Ng5 3.fxg5 Kf7 4.gxf6 Rg8 5.Qh3 Rh8 6.Qg4 Qxf6 7.Rb1 Rhb8 8.Re1 Rb7 9.Bd3 Be6 10.Qh5+ Kg8 11.Rf1
  +-  (3.85)   Depth: 17/40   00:00:13  92514kN
1.Rxg7+
  +-  (4.10)   Depth: 18/40   00:00:20  135mN
1.Rxg7+ Kxg7 2.Qg3+ Ng5 3.fxg5 Kf7 4.Qf4 Rh5 5.c4 Bxg2 6.Bxg2 Rxg5 7.Bd4 Ra6 8.h4 Rg8 9.Rf1 c5 10.Bxc5 Rxa4 11.Qe4 Qd7 12.Qh7+ Rg7
  +-  (4.31)   Depth: 18/44   00:00:26  175mN
1.Rxg7+ Kxg7 2.Qg3+ Ng5 3.fxg5 Kf7 4.Qf4 Rg8 5.c4 Bxg2 6.Bxg2 Ra5 7.Qf5 Rxg5 8.Qh7+ Rg7 9.Qh5+ Kg8 10.Rd1 Qb8 11.Re1 Ra8 12.Bd4 Rg5 13.Qh6 Rxa4 14.Bxf6
  +-  (4.17)   Depth: 19/46   00:00:51  345mN
1.Rxg7+
  +-  (4.42)   Depth: 20/46   00:01:08  460mN
1.Rxg7+
  +-  (4.92)   Depth: 20/48   00:01:26  575mN

(, Prakash 02.03.2010)

It took 12 seconds to find it and x 4 times time to confirm it. Great.
Parent - - By turbojuice1122 (Gold) Date 2010-03-02 03:49
On the first one, I'm glad that the move and continuation that I found in about two seconds turned out to be the right one. :-)
Parent - - By Master Om (Bronze) Date 2010-03-02 03:52
you found it manually ? Great ?
Parent - - By turbojuice1122 (Gold) Date 2010-03-02 03:55
Take it back--I was looking more at the situation where the knight goes to d5, not to g4.  Even so, though I still haven't checked it out on a machine, I would still be interested in sacrificing the bishop on h7 in the Ng4 case, too.  It seems like at the very least, white gets a free pawn and an open kingside.
Parent - By Master Om (Bronze) Date 2010-03-02 03:57
It needs to run for more time. It was only 2 seconds.
Parent - By yanquis1972 (****) Date 2010-03-02 03:51
from the test positions & solutions i've seen posted on here & talkchess, i would say one-thousand-times yes. it's quite astonishing tactically.
Parent - - By yanquis1972 (****) Date 2010-03-02 03:39
fritz slower than shredder - unbelievable! like you say, not a knock on shredder, which is great, but it's a snail tactically. never thought i'd see the day when fritz lags behind it in tactics. wow.

great review & much appreciated. i don't think i'll be purchasing the engine, despite my addiction, particularly with personal faves hiarcs & rybka surely around the corner.  fritz the engine has never seemed to offer anything particularly special since its days as top dog vanished. once it stopped outrunning the engines, it never developed any particular strength or style to make it appealing...i have fritz 12 SP & find it massively disappointing.  i was also astonished at some of its tactical inability even in my limited play with it.  i was running thru some of the games of the infamous computer cheater clemens allwermann with Fritz 5.32 & discovered a few great king sacs & tactical shots he missed for whatever reason -- on one in particular -- Fritz 12 was something like 2-3 minutes slower.   i realize fritz 5.32 is regarded as a tactical monster relative to humans, but i'd at least guess it's safe to say it's hugely outclassed by virtually all of today's top engines given search advancements. that was just one position, but it suprised me quite a bit.
Parent - By BrandonSi (***) Date 2010-03-02 03:56
"i don't think i'll be purchasing the engine, despite my addiction, particularly with personal faves hiarcs & rybka surely around the corner."

I'm with you 100%. The day HIARCS 13 is announced is the day they get my money. Obviously the same with Rybka 4, but unfortunately I feel that day might be a long ways off. At least the "Rybka" money I have set aside is earning interest while we wait! :)
Parent - - By FWCC (***) Date 2010-03-03 22:00
Did you put Stockfish to the test in your testsuites?
Parent - - By Carl Bicknell (*****) Date 2010-03-04 00:07

> Did you put Stockfish to the test in your testsuites?


The old versions did OK, I haven't tried the latest stockfish - will test in due course!
Parent - By Uly (Gold) Date 2010-03-04 00:16
Don't test the latest, I think the version we're interested in is 1.6s ;) (the good one).
Parent - By M ANSARI (*****) Date 2010-03-04 07:12
Although I bought this GUI I have not had time to test it.  Someone mentioned that it had fixed the R3 EGTB bug.  Could you test to see that?
Up Topic The Rybka Lounge / Computer Chess / Deep Fritz 12 mini review

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill