Author: Sven Reichard
Date: 02:14:25 06/24/02
Go up one level in this thread
Folks,
here's the second part of Eduard's article in English. I won't
translate Fritz' output; remember that the piece letters are
English K Q B N R (P)
German K D L S T (B)
>
>Test position 1 (Mate in 2)
>
>
>
>[D]8/4B3/5K1p/5p1k/5Pb1/5p1p/5P1P/8 w - - 0 1
>
>Analysis by Fritz 6c:
>
>1.La3
> = (0.09) Tiefe: 25/28 00:00:03 1450kN
>1.La3!
> ² (0.41) Tiefe: 26/34 00:00:06 3071kN, tb=3
>
>
>Analysis by Fritz 7:
>
>1.Ld8!
> +- (#2) Tiefe: 5/5 00:00:00
>
We see that whereas Fritz 6 can't find the Mate in 2 due to null move
problems, Fritz 7.001 doesn't have any problems.
>
>
>[D]8/8/p3R3/1p5p/1P5p/6rp/5K1p/7k w - - 0 1
>
>Analysis by Fritz 7.001:
>
>1.Txa6 Tg2+ 2.Kf3 Tg8 3.Ta1+ Tg1 4.Ta6 Tf1+ 5.Ke3 Tb1 6.Kf2
> -+ (-3.87) Tiefe: 12/21 00:00:00 290kN
>1.Te1+!
> +- (#7) Tiefe: 12/21 00:00:00 290kN
>1.Te1+!
> +- (#7) Tiefe: 12/21 00:00:00 305kN
>
And what's Fritz 6c's analysis?
>Analysis by Fritz 6:
>
> 1.Txa6 Tg2+ 2.Kf3 Tg5 3.Ta1+ Tg1 4.Ta8 Tf1+ 5.Ke3 Tf7
> -+ (-4.59) Tiefe: 16/31 00:00:07 3990kN
>1.Txa6 Tg2+ 2.Kf3 Tg5 3.Ta1+ Tg1 4.Ta8 Tf1+ 5.Ke3 Tf7
> -+ (-4.72) Tiefe: 17/30 00:00:12 6425kN
>1.Txa6 Tg2+ 2.Kf3 Tg5 3.Ta1+ Tg1 4.Ta8 Tf1+ 5.Ke3 Tf7
> -+ (-4.81) Tiefe: 18/33 00:00:17 9694kN
>
Frizt 6 isn't able to find the the mate.
Before playing the new engine myself I couldn't help checking the
following tactical position with Fritz 7: It stems from Kasparov's
analysis of his WC match with Karpov. Kasparov probably could have won
the game with Rxe6!
>
Here Fritz 6's analysis:
>
>[D]r4rk1/p1q2ppp/1p1Nn3/8/3R2Q1/1P4P1/P4PKP/4R3 w - - 0 1
>
>Analysis by Fritz 6c:
>
>1.Td5 Dc6 2.Ted1 Tad8 3.Df3 a6 4.T5d2 Dxf3+ 5.Kxf3 Td7
> ² (0.38) Tiefe: 14/29 00:01:36 41203kN
>1.Txe6
> ² (0.41) Tiefe: 14/36 00:02:41 69592kN
>1.Txe6 h5 2.De4 fxe6 3.Dxe6+ Kh7 4.Tc4 Dd8 5.De4+ Kh8
> ± (0.88) Tiefe: 14/37 00:03:48 98728kN
>
Can Fritz 7.001 hold its own tactically?
>Analysis by Fritz 7:
>
>
>1.Td5 Dc6 2.Dc4 Dxc4 3.bxc4 Tfd8 4.Ted1 a6 5.h3 Sc7 6.T5d2 Se8 7.Sxe8
> ² (0.54) Tiefe: 14/35 00:02:38 57716kN
>1.Txe6
> ² (0.57) Tiefe: 14/35 00:03:27 75978kN
>
In spite of the lower raw speed, Fritz 7.001 succeeds in playing not
substantially worse than the last version of Fritz 6!
In the area of King's attack, Fritz 7's engine was even improved
significantly, which is also proven with the new positional test suite
(WM-Test) by Dr. M. Gurevich and H.J.S. Fritz 7.006 currently (05/02)
has the lead in that test area.
Here is a small example. Even on my slow P600, Fritz 7.006 plays the
winning move within seconds.
>
>
>Soffer,R - Fritz 6
>
>[D]r3qrk1/4bppp/4p3/p2pP2Q/1p1B4/1PpPP3/P1P2RPP/5RK1 w - - 0 1
>
>Analysis by Fritz 7:
>
>21.Tf6
> ² (0.54) Tiefe: 12/30 00:00:12 3881kN
>21.Tf6
> ² (0.69) Tiefe: 12/30 00:00:14 4871kN
>21.Tf6!
> ± (0.97) Tiefe: 13/33 00:00:29 10051kN
>21.Tf6!
> ± (1.25) Tiefe: 13/33 00:00:38 12890kN
>21.Tf6!
> +- (1.54) Tiefe: 14/33 00:01:00 20292kN
>21.Tf6!
> +- (1.82) Tiefe: 14/35 00:01:18 26533kN
>21.Tf6!
> +- (2.38) Tiefe: 14/35 00:01:37 33079kN
>
>Soffer,R - Fritz 6
>r3qrk1/4bppp/4p3/p2pP2Q/1p1B4/1PpPP3/P1P2RPP/5RK1 w - - 0 1
>
>Analysis by Fritz6c:
>
>21.Tf6
> = (0.22) Tiefe: 14/36 00:00:57 25007kN
>
As an aside: GM Ram Soffer was present at the Fritz 7 server during
the last game of Dj vs. GM Ilya Smirin and commented on it. I recall
his words well. When the game draw to its end, Ram Soffer said
something like: White's pieces are very actively placed, hence it's
probably going to be a draw. And that's what happened.
Hence, programmer Frans Morsch succeeded in keeping the tactical power
of the new engine about at the same level, and at the same time
improved it in other areas, which already could be seen in the
zugzwang positions. According to Morsch also its endgame play has been
improved; likewise, pawn structures have been totally reworked.
My first game with Fritz 7.001
>Ich war auf meine erste Partie gegen Fritz 7 sehr neugierig, insbesondere wollte
>ich wissen wie die neue Engine mit Trojanerstellungen umzugehen weiss. Zu meiner
>Überraschung
>gelang es mir, Fritz 7 mit einigen Trojanerstellungen total zu überspielen,
>schneller sogar als früher Fritz 5.32!
I was quite curious about my first game with Fritz 7; in particular I
wanted to know how the new engine deals with Trojan positions. To my
great surprise I managed to beat Fritz 7 convincingly with some Trojan
positions, even faster than it had been with Fritz 5.32.
Here is the very first game (I played a few):
>
>[Event "Blitz10'"]
>[Site "?"]
>[Date "2001.??.??"]
>[Round "?"]
>[White "Nemeth,Eduard"]
>[Black "Fritz 7.001"]
>[ECO "A00"]
>[Result "1-0"]
>
>1. d3 {0} d5 {0} 2. c3 {4} e5 {-0.25/12 27} 3. g3 {6} Nf6
>{0} 4. h4 {5} Bd6 {-0.66/11 32} 5. Bg2 {9} O-O {-0.69/11
>14} 6. Nh3 {9} c6 {-0.91/9 14} 7. Bg5 {12} h6 {-1.16/10 18}
>8. e4 {8} hxg5 {-2.00/9 5} 9. hxg5 {3} Bg4 {-1.97/9 9}
>10. f3 {10} Bxh3 {-2.22/9 11} 11. Rxh3 {12} Nfd7 {-2.04/9
>17} 12. f4 {3} Qb6 {-1.72/8 25} 13. Qh5 {5} Qe3+ {0.00/7
>12} 14. Kf1 {6} f6 {5.06/8 27} 15. g6 {2} Qc1+ {8.90/9 29}
>16. Ke2 {31} Qd2+ {#6/7 1} 17. Kxd2 {6} dxe4 {#1/2 6}
>18. Qh8# {2} 1-0
>
In this game the following position arose:
>
>[D]rn3rk1/pp1n1pp1/1qpb4/3pp1P1/4PP2/2PP2PR/PP4B1/RN1QK3 w Q - 0 1
>
This position is easily won for White after Qh5! There is no
defense. Why was Fritz 7 taken off guard? An explanation could be a
bug in the engine - but I don't know. Even in this won position Fritz
7.001 shows a positive evaluation for black (in spite of the right
white move), and this evaluation didn't change for 15 min.
>Analysis by Fritz 7.001:
>
>1.Dh5
> -+ (-1.75) Tiefe: 8/39 00:00:47 12645kN
>1.Dh5 Dg1+ 2.Lf1 Dxf1+ 3.Kxf1
> -+ (-1.71) Tiefe: 8/46 00:00:55 14660kN
>
I posted the game including Fritz' analysis at the usual computer
chess boards, with the conclusion "this is humbug" (sorry, didn't find
a more adequate expression, EN).
After only a few weeks ChessBase published an update for Fritz 7
including a new engine (Fritz 7.002, 424 kB)!
The version number of the engine can be found by right-clicking on the
engine "Fritz 7" in the folder ChessBase\Engines, and then selecting
"Properties"!
Then, when I tested this engine, I couldn't believe my eyes! In a
matter of 3 weeks the programmer managed to improve the engine
dramatically - some people estimated the difference to Fritz 7.001 as
something like 100 Elo!. Here is the same position, analyzed by Fritz 7.002:
>Analysis by Fritz 7.002:
>13.Dh5!
>+- (5.35) Tiefe: 8/38 00:00:17 4828kN
>13.Dh5 De3+ 14.Kf1 Dxd3+ 15.Kg1 De3+ 16.Kh2 Dxg3+ 17.Kxg3 exf4+ 18.Kf2 Lc5+
>19.Kf1 f6 20.Dh7+
>+- (5.79) Tiefe: 8/40 00:00:20 5616kN
>
This new Fritz 7 engine was very convincing and took the lead in
private rating list, e.g., Gerhard Sonnabend's (SSDF hadn't tested
Fritz 7 at that point).
I wanted to test the new Fritz 7.002 engine myself, and since I had
developed a new middle game test I was happy to get the new engine.
Fritz 7.002 convinced me in the endgame, too. In my "Middle game test
without queens", Shredder 5.32 basically got its butt kicked, and
Chess Tiger 14.0 lost by a small margin. Many people agreed at that
point: The new endgame reference is Fritz 7!
Fritz 7.002 at times even played moves that are totally atypical for
computers (they appear rather human), as the following position illustrates:
>Fritz 7.002 - Gambit Tiger 2.0
>
>[D]6k1/p1q2ppb/Q4n1p/2N5/3P4/P4RN1/1r4PP/6K1 w - - 0 1
>
>
>Analysis by Fritz 7.002:
>
>32.Txf6 gxf6 33.Dxf6 Tb1+ 34.Kf2 Da5 35.De5 Dd2+ 36.Kf3 Lg6 37.Se2 Tf1+
>³ (-0.28) Tiefe: 12/35 00:00:40 13699kN
>32.Txf6 gxf6 33.Dxf6 Tb1+ 34.Kf2 Lg6 35.Kf3 Tb2 36.d5 Tc2 37.Sce4 Lxe4+
>³ (-0.40) Tiefe: 13/38 00:02:12 44660kN
>
Here, Fritz 7.002 courageously sacrifices the exchange, which maybe
isn't optimal against his peers, but when playing humans it certainly
would surprise the opponent. The position remains complicated, but for
programs - in particular for Fritz - it's just what they like!
>Bug in the Fritz 7.002 engine
>
My little middle game test got feedback from a couple of testers. In
one game the following position arose after move 13 (see diagram):
>Gandalf_432h - Fritz 7.002
>
>1.e4 c5 2.g3 d5 3.exd5 Dxd5 4.Df3 Dxf3 5.Sxf3 Lf5 6.Lg2 Sc6 7.d3 g6 8.Le3 Lg7
>9.Sc3 c4 10.Sh4 Le6 11.d4 Sxd4 12.Lxb7 Tb8 13.Le4
>
>[D]1r2k1nr/p3ppbp/4b1p1/8/2pnB2N/2N1B1P1/PPP2P1P/R3K2R b KQk - 0 13
>
>13. ... Txb2? 14.0-0-0 Se2+ 15.Sxe2 Ld7 16.Ld5 Txa2 17.Lxc4 Ta5 18.Sf3 Lc6
>19.Sed4 Lxd4 20.Lxd4 f6 21.Le2 e5 22.Lc3 Td5 23.Txd5 Lxd5 24.Td1 Se7 25.Lb4 Le6
>26.Lc5 Kf7 27.Td6 Tc8 28.Lxa7 Sc6 29.Lc5 Se7 30.Lb4 Sd5 31.La5 Sc3 32.Lxc3 Txc3
>33.Ld3 Ke7 34.Ta6 Tc7 35.Le4 Tc4 36.Ta7+ Kd6 37.Ld3 Tc7 38.Txc7 Kxc7 39.Kb2 Kd6
>40.Kc3 Kc5 41.Sd2 f5 42.Sb3+ Kd6 43.Kb4 h6 44.c4 Kc7 45.Lc2 Lc8 46.Kc5 Le6
>47.Sd2 g5 48.Sf3 e4 49.Sd4 Lc8 50.Kd5 g4 51.Ke5 h5 52.Sxf5 Lb7 53.Lxe4 La6
>54.Se3 Kd7 55.Lg6 1-0
>
After Fritz 7.002 played 13...Rxb2, Gandalf 4.32h simply won the
exchange by queenside castling!
People tried to guess the reason. In the meantime we know a bit more
about it. In most cases where queenside castling is possible, Fritz
7.002 considers one move less than all the possible ones. I could
produce numerous examples. Using the multiple variation mode it can be
proven easily; here is a simple example:
>[D]8/8/8/8/2p5/2pkp1N1/8/R3K2R w KQ - 0 1
>
White to move and mate in 1!
And here the multiple variation analysis (!) by Fritz 7.002:
>1. +- (#6): 1.Td1+
>2. +- (#9): 1.Se2
>3. +- (#10): 1.Th4
>4. +- (#12): 1.Ta8
>5. +- (#14): 1.Th6
>6. +- (#11): 1.Th8
>7. +- (#11): 1.Th7
>8. +- (#12): 1.Th5
>9. +- (#12): 1.Ta7
>10. +- (#13): 1.Ta3
>11. +- (15.57): 1.Ta5
>12. +- (13.44): 1.Th3
>13. +- (13.41): 1.Tc1
>14. +- (13.41): 1.Th2
>15. +- (13.35): 1.Tb1
>16. +- (13.19): 1.Tf1
>17. +- (13.19): 1.Tg1
>18. +- (13.19): 1.Ta4
>19. +- (13.13): 1.Sf1
>20. +- (13.07): 1.Ta2
>21. +- (13.07): 1.Ta6
>22. +- (12.79): 1.Kd1
>23. +- (12.60): 1.Kf1
>24. +- (12.54): 1.Sf5
>25. +- (12.47): 1.Sh5
>26. +- (11.00): 1.0-0
>27. +- (10.04): 1.Se4
>
Fritz 7.002 recognizes only 27 possible moves... however there are 28,
and the best move (O-O-O#) is missing. What now?
Hence, there had to be a new update, as such a bug couldn't be left
in. For a serious tournament player such an engine has only a limited
use, since queenside castling occurs in a lot of games.
The next Fritz 7 update I remeber as follows: On 01/07/02 I was online
at the F7 server Schach.de for a couple of hours, when suddenly a new
update was announces. After the setup I immediately looked into the
folder ChessBase\Engines, where I found a new engine dated
01/07/02. The version number was 7.006!
At this point I would like to list all Fritz 7 engines published so
far, in order to not lose track in what follows.
>Fritz 7.001 = Datum 13.10.01 (420 kB)
>Fritz 7.002 = Datum 05.11.01 (424 kB)
>Fritz 7.006 Beta = Datum 07.01.02 (424 kB)
>Fritz 7.006 = Datum 11.01.02 (424 kB)
>Fritz 7.007 = Datum 23.03.02 (424 kB)
>
As somebody related to the message board, the version released on
01/07/02 was a 7.006 beta version that went into the update by
mistake. The update was available on the Fritz 7 server only for a
couple of hours, hence not everybody has this new engine. A couple of
days later there was a new update with the final version of the Fritz
7.006 engine - date: 01/11/02.
Needless to say that the castling bug was eliminated in version
7.006. We, the users, were happy, and the testing could start anew.
I myself started by checking the difference between the beta version
and the version at 01/11/02.
And - I really could see a few differences. Even Professor E. Irazoqui
posted two positions on CCC that showed differences between the two
versions.
Here is the first one:
>
>[D]8/1p6/p4p1p/2p1P2k/5P2/6pP/1P4P1/6K1 w - - 0 1
>
>1.f5!!
>
>The second one:
>
>[D]2r1r1k1/p4pp1/1p1Pn1p1/6q1/P5P1/BP2Pp1P/5P2/Q1RR2K1 b - - 0 1
>
>1. Sf4!!
>
In both positions the strange thing was that the beta engine found the
solution faster than the release engine.
A very interesting analysis happens in the following position:
>[D]r1b2rk1/pp2bpp1/1qn1pn1p/6B1/2PpP2P/3P2PN/P1P2PB1/RN1Q1K1R b - - 0 1
>
>Analysis by Fritz 7.006 (Beta vom 07.01.02):
>
>12...hxg5 13.hxg5 Sg4 14.Sd2 Sce5 15.Sb3 Lb4 16.Tb1 Td8 17.g6 fxg6 18.Kg1
> -+ (-2.19) Tiefe: 12/42 00:00:13 3172kN
>12...hxg5 13.hxg5 Sg4 14.Sd2 Sce5 15.Sb3 Lb4 16.Tb1 f5 17.g6 Sxg6 18.Sxd4 Dd6
> -+ (-2.04) Tiefe: 13/41 00:00:29 8053kN
>
>Analysis by Fritz 7.006 (Datum: 11.01.02 - offizielle Version!):
>
>12...Db2 13.e5 Dxa1 14.exf6 Lxf6 15.Lxf6 gxf6 16.Lxc6 bxc6 17.Dg4+ Kh7 18.Kg2 f5
>19.Df3 Dxa2
> -+ (-2.13) Tiefe: 12/30 00:00:10 2740kN
>12...Db2 13.Sd2 hxg5 14.hxg5 Sd7 15.Sb3 g6 16.Kg1 Sb4 17.Tb1 Dxc2 18.Dxc2 Sxc2
>19.a4
> -+ (-2.06) Tiefe: 13/36 00:00:26 7324kN
>
Well - the programmer apparently implemented some more knowledge of
Trojan positions!
The version of 01/11/02 is the official Fritz 7.006 which plays also
at SSDF.
In March another update contained yet another Fritz 7 engine, version
number 7.007. This engine is identical with 7.006 as far as the chess
code is concerned. The only change is the size of the hash table
management. Fritz 7.006 can't deal with hash tables greater than 1GB,
whereas the newest version does that - so that the absolute freaks
with 2GB of RAM don't complain!
I already mentioned that Fritz 7 was considerably improved in the
endgame. In conclusion let us look at positions where Fritz
understands what's going on even without table bases:
>[D]7k/1K5P/6B1/1P6/3b4/8/8/8 w - - 0 1
>
>(all analyses w/o table bases)
>
>Analysis by Shredder 5.32:
>
>1.Kc6 Kg7 2.Le4 Lf2 3.Kd7 Lg1 4.h8D+ Kxh8 5.Ke8
>+- (2.85) Tiefe: 18/30 00:00:08 2275kN
>1.Kc6 Kg7 2.Lf5 La7 3.Kd7 Ld4 4.h8D+ Kxh8 5.Ke6 Lf2 6.Lh7 Kxh7 7.Kf6
>+- (2.84) Tiefe: 19/31 00:00:12 3275kN
>1.Kc6 Kg7 2.Le4 Lf2 3.Kd7 Lg1 4.h8D+ Kxh8 5.Ke8 Kg7 6.Ld5 Kh8 7.b6
>+- (2.84) Tiefe: 20/32 00:00:16 4534kN
>
>Analysis by Chess Tiger 14.0:
>
>1.Kc6 Kg7 2.Le4 La7 3.Lf5 Le3 4.Kd5 Lf2 5.Lc2 Lb6 6.Ld3 La7 7.Kc6 Ld4 8.Lg6 La7
>9.Le4
>+- (2.09) Tiefe: 17 00:00:09 2513kN
>1.Kc6 Kg7 2.Le4 La7 3.Lf5 Lf2 4.Ld3 La7 5.Le4 Kh8 6.Kc7 Kg7 7.Lc2 Lc5 8.Kc6 La7
>9.Ld3 Ld4
>+- (2.09) Tiefe: 18 00:00:25 7431kN
>
>And now Fritz 7:
>
>Analysis by Fritz 7 (directly from the game, for this position is not
>artificial)
>
>89.b6 Le5
>= (0.00) Tiefe: 23/34 00:00:22 11653kN
>89.b6 Le5
>= (0.00) Tiefe: 24/36 00:00:36 18876kN
>
Fritz 7 is the only program that evaluates this position as 0.00,
hence better than endgame virtuoso Shredder 5.32 (even Shredder 6.02
w/o table bases evaluates this as +3 for White).
Frans Morsch managed to improve the new engine significantly, and it
may be assumed that this will continue for a while. Let's therefore be
happy about the new Fritz 7 engine that currently leads all ranking
lists, and, as I showed above, deservedly so.
Appendix:
On May 24 there was an inofficial update on the F7 server, including a
new engine (dated 05/19/02).
I posted the following game on May 18 on CCC. Fritz plays for
threefold repetition, although his score is +2.13 ??
>Fritz 7.0.0.6 - Nemeth,E [C57]
>Blitz=60' Stuttgart, 18.05. 2002
>[Nemeth,Eduard]
>1.Sf3 0 1...Sc6 3 2.e4 0 2...e5 1 3.Lc4 0 3...Sf6 2 4.Sg5 0 4...d5 1 5.exd5 0
>5...Lg4 8 6.Sxf7 0.94/13 76 6...Dc8 4 7.f3 1.97/14 141 7...Sd4 11 8.Sxh8 4.09/13
>87 8...Lxf3 4 9.gxf3 3.88/12 259 9...Dh3 8 10.Tf1 3.28/11 136 10...Dh4+ 19
>11.Tf2 3.03/12 0 11...Lc5 15 12.Kf1 2.09/12 116 12...Dh3+ 44 13.Ke1 1.84/12 23
>13...Dh4 38 14.Kf1 2.13/12 87 14...Dh3+ 20 15.Ke1 1.75/13 67 15...Dh4 13 16.Kf1
>2.13/12 60 ½-½
>
>Nach Zug 15...Dh4:
>
>[D]r3k2N/ppp3pp/5n2/2bPp3/2Bn3q/5P2/PPPP1R1P/RNBQK3 w q - 0 1
>
>Analysis by Fritz 7.0.0.6 (F7.0.0.7 ist identisch):
>
>16.Kf1 Sb3 17.Lb5+ Kf8 18.d4 Lxd4 19.Sg6+ hxg6
> +- (4.22) Tiefe: 6/23 00:00:00 69kN
>16.Kf1 Sb3 17.Lb5+ Kd8 18.d4 Lxd4 19.Le3 Sxa1
> +- (4.09) Tiefe: 7/26 00:00:00 152kN
>16.Kf1--
> +- (3.81) Tiefe: 8/21 00:00:01 269kN
>16.Kf1 Sb3 17.Tg2 Sxa1 18.De2 Ld6 19.Lb5+ Kf8 20.c4
> +- (3.28) Tiefe: 8/34 00:00:01 470kN
>16.Kf1 Sb3 17.Tg2 Sxa1 18.De2 e4 19.Lb5+ Kf8 20.Dc4 Ld6 21.fxe4
> +- (3.25) Tiefe: 9/33 00:00:04 1246kN
>16.Kf1--
> +- (2.97) Tiefe: 10/36 00:00:08 2439kN
>16.Kf1
> +- (2.97) Tiefe: 10/36 00:00:12 3564kN
>16.Kf1--
> +- (2.69) Tiefe: 11/36 00:00:17 4921kN
>16.Kf1
> +- (2.69) Tiefe: 11/40 00:00:25 7792kN
>
>(Nemeth, Stuttgart 25.05.2002)
>
>Fritz 7 - Nemeth,E
>r3k2N/ppp3pp/5n2/2bPp3/2Bn3q/5P2/PPPP1R1P/RNBQK3 w q - 0 1
>
>Analysis by Fritz 7.0.0.8:
>
>16.d3 Sb3 17.Lb5+ c6 18.d4 Lxd4 19.Lxc6+ bxc6
> +- (3.09) Tiefe: 6/23 00:00:00 92kN
>16.d3 Sb3 17.Lb5+ Sd7 18.Lxd7+ Kxd7 19.d4 Lxd4 20.Le3 Sxa1
> +- (2.97) Tiefe: 7/23 00:00:00 129kN
>16.d3--
> +- (2.69) Tiefe: 8/24 00:00:00 221kN
>16.d3 Sb3 17.De2 Sxa1 18.Dxe5+ Kf8 19.Dg3 Lxf2+ 20.Kxf2 Dxg3+ 21.hxg3
> +- (2.13) Tiefe: 8/35 00:00:01 353kN
>16.d3 Sb3 17.De2 Sxa1 18.Dxe5+ Kf8 19.Dg3 Lxf2+ 20.Kxf2 Dxg3+
> +- (2.13) Tiefe: 9/29 00:00:03 846kN
>16.d3--
> +- (1.84) Tiefe: 10/30 00:00:05 1488kN
>16.d3
> +- (1.84) Tiefe: 10/35 00:00:07 2136kN
>16.d3 Sb3 17.De2 Lxf2+ 18.Dxf2 Dxf2+ 19.Kxf2 Sxa1 20.d6 cxd6 21.Sf7 Ke7
> +- (1.75) Tiefe: 11/36 00:00:18 5778kN
>
>(Nemeth, Stuttgart 25.05.2002)
>
>Wie man sieht, zieht Fritz 7.0.0.8 hier d3!? Das ist korrekt ist (+=).
As you see, Fritz 7.0.0.8 moves d3!? (don't understand the rest - transl.)
>
The new engine 7.0.0.8 is dated 5/19/02 (11:51 p.m.!): So Frans reads
>my postings after all.
IMPORTANT!
To detect the fine analytical difference you have to save the whole
game and click on Black's move 15...Qh4 - FEN alone is NOT sufficient.
Fritz 7.0.0.6 (7007 as well) plays then Kf1? again. The new engine
would play 16. d3 - thereby evading repetition. When the whole game is
loaded, 7008 apparently the repetition threat. With FEN alone that
can't be determined. Thus: Please don't test it like that!
Here is the proof again, posted like that on CCC on June 23:
I played the following game with Fritz 7.006 (published here), when I
was surprised seeing Fritz with +2 go for some weird repetition! :(
Draw at move 16:
>
>[Event "Blitz=60' Stuttgart [Nemeth,Eduard]"]
>[Site "?"]
>[Date "2002.??.??"]
>[Round "?"]
>[White "Fritz 7006"]
>[Black "Nemeth,Eduard"]
>[ECO "C57"]
>[Result "1/2-1/2"]
>1. Nf3 {0} Nc6 {3} 2. e4 {0} e5 {1} 3. Bc4 {0} Nf6 {2}
>4. Ng5 {0} d5 {1} 5. exd5 {0} Bg4 {8} 6. Nxf7 {0.94/13 76}
>Qc8 {4} 7. f3 {1.97/14 141} Nd4 {11} 8. Nxh8 {4.09/13 87}
>Bxf3 {4} 9. gxf3 {3.88/12 259} Qh3 {8} 10. Rf1 {3.28/11
>136} Qh4+ {19} 11. Rf2 {3.03/12 0} Bc5 {15} 12. Kf1
>{2.09/12 116} Qh3+ {44} 13. Ke1 {1.84/12 23} Qh4 {38}
>14. Kf1 {2.13/12 87} Qh3+ {20} 15. Ke1 {1.75/13 67} Qh4
>{13} 16. Kf1 {2.13/12 60} 1/2-1/2
>
Now I saved the most important moves from this game in a "test book"
and let Fritz 7.008 try the position. In the beginning F7008 really
played 14.Kf1?, but later it changed to the much better 14. d3!!
Here is the proof:
>
>[Event "Bugbeseitigung [Nemeth,Eduard]"]
>[Site "?"]
>[Date "2002.??.??"]
>[Round "?"]
>[White "Fritz 7008"]
>[Black "Nemeth,Eduard"]
>[ECO "C57"]
>[Result "*"]
>1. Nf3 {0} Nc6 {2} 2. e4 {0} e5 {3} 3. Bc4 {0} Nf6 {2}
>4. Ng5 {0} d5 {1} 5. exd5 {0} Bg4 {3} 6. Nxf7 {0} Qc8 {3}
>7. f3 {0} Nd4 {2} 8. Nxh8 {0} Bxf3 {1} 9. gxf3 {0} Qh3 {2}
>10. Rf1 {0} Qh4+ {3} 11. Rf2 {3.84/9 0} Bc5 {9} 12. Kf1
>{2.69/11 45} Qh3+ {8} 13. Ke1 {1.75/13 87} Qh4 {6} 14. d3
>{1.81/12 66} Nb3 {24} 15. Qe2 {1.34/12 28} *
>
Well done, Frans Morsch (Congratulations!). This can't be detected
using test position suites; there F7008 gives the same results as
F7006. Fortunately there are still players that play themselves,
otherwise this bug could have gone undetected forever!
By the way: This bug occured already in Fritz 6!!
Both the game above and the one below against Fritz 6c I played on a
P600, 128 MB hash at G/30.
>
>Fritz6c - Nemeth,Eduard
>128MB, test.ctg, P600
>1.Sf3 Sc6 2.e4 e5 3.Lc4 Sf6 4.Sg5 d5 5.exd5 Lg4 6.Sxf7 Dc8 7.f3 Sd4 8.Sxh8 Lxf3
>9.gxf3 Dh3 10.Tf1 Dh4+ 11.Tf2 3.88/9 0 Lc5 12.Kf1 2.31/11 41 Dh3+ 13.Ke1 2.16/12
>36 Dh4 14.Kf1 2.25/12 1:11 Dh3+ 15.Ke1 2.13/13 1:29 Dh4 16.Kf1 2.25/12 1:07 ½-½
>
It is identical with the first one against Fritz 7.006.
Hence it is seen here that this strange bug existed already in earlier
version of Fritz - and it was never detected - until recently!!!
>Ich hoffe dass mir wenigstens Frans da ein wenig dankbar ist (andere sind's
>nicht) dass er seinen Fritz verbessern konnte. Gut hast Du das gemacht Frans mit
>Fritz 7.008. Es ist wohl daher die beste Engine derzeit, sowohl für das Spielen
>als auch für die Analyse. Weiter so Frans...ich werde am Ball bleiben!
I hope that at least Frans is a little bit grateful (others aren't)
that he could improve his Fritz. Well done, Frans, with Fritz
7.008. Hence it is currently the best engine, both for playing and for
analysis. Way to go, Frans... I'll be there!
>
>Eduard
>
PS: Best regards to Frans Morsch!
>
>Eduard Nemeth, 23.06.2002.
Translation: Sven Reichard, 06/24/02.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.