Author: Moritz Berger
Date: 10:34:42 05/17/98
Go up one level in this thread
On Sun, 17 May 1998 10:01:19 GMT, DFrickenschmidt@w-i-s.net (Dirk
Frickenschmidt) wrote:
>In CBM 63, in the 'Fritz5 forum', I found an interesting position
>illustrating problems of king security.
>
>Fritz5,P (2500) - Golubev,M (2520) [E98]
>Sparkassen Open rapid (7), 1998
>
>1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5 7.0-0 Nc6 8.d5
>Ne7 9.Ne1 c5 10.dxc6 bxc6 11.b4 d5 12.Bg5 d4 13.Bxf6 Bxf6 14.Na4 Bg7
>15.Nc5 h5 16.f4 exf4 17.Ned3 h4 18.Rxf4 a5 19.b5 cxb5 20.cxb5 a4
>21.Rb1 Bh6 22.Rf1 Be3+ 23.Kh1 f5 24.Bf3 fxe4 25.Bxe4 Nf5
>
>The critical position.
>On rapid chess level Fritz5 did not see the danger for white fast
>enough and lost the game.
>
>I had a look at this position on a Pentium 200MMX/48MbHash.
>
>*Fritz5* needs about 7 minutes to 'smell' the danger of being mated
>and switches to 26. b6
>This looks sufficient to me to keep a decisive advantage.
>Black seems to find no hole to cause big white problems.
Here's the position in EPD format:
r1bq1rk1/8/6p1/1PN2n2/p2pB2p/3Nb3/P5PP/1R1Q1R1K w - -
Solution times so far (ranking according to times to play a different
move than Bxa8, details below, all times on a P233MMX, 128MB RAM):
Program / Ne5 / b6 / g4
-----------------------------------
1. MCP 7.1 / - / 12:27 / 03:29
2. Fritz 5 / - / 05:39 / 24:54
3. Hiarcs 6 / - / 10:57 / 15:44
4. Shredder 2 / 21:47 / 01:09:48 / 01:53:44
5. Rebel 9 / 35:54 / 51:44 / not tested
6. CSTal Paris / - / 01:20:46 / 02:08:42
I tried the position with Fritz 5 on P233MMX with 98304 KB hash tables
The problem for Fritz here is move ordering on 12th ply search depth,
which it needs to spot the trouble:
the first 12ply PV 26.Bxa8 shows up after 49" (hash tables 45% full)
26.b6 is 4th best moves at that depth and is confirmed better after 5:39
(eval +- 1.09, hash tables full)
after 11:37 we get a full PV
26.b6 Ng3+ 27.hxg3 Rxf1+ 28.Qxf1 hxg3 29.Nf4 Qh4+ 30.Nh3 Bxh3
(score +- 2.59)
The score for 26.b6 is so high that it takes even longer (still on
12ply) to find out that 26.g4 (move 28 of 45) is the real winner:
+- 2.63 after 24:54 (24 minutes: 54 seconds, >463 million nodes
searched), no PV other than 26.g4 given (= confirmed fail high).
M-Chess Professional 7.1 gets it right at 8ply already, chosing 26.b6
over 26.g4 after some time (it clearly plays the moves for positional
reasons after about 20 million nodes and doesn't have the 20 times
bigger search tree to see that 26.g4 is eventually even stronger):
8 ply
03:29 planning 26.g4 (+3.29)
12:27 planning 26.b6 (+3.69)
A really nice result in this tough position.
Rebel 9
11.00 ply 18:11 Bxa8 fails low, score +0.63 (was +4.03 at 10 ply)
- after 25 minutes (about 110 million nodes) q-search peak at 57 ply -
11.01 ply 35:54 Ne5 Ra7 Nxg6 Rf6 Ne5 Qd6 Nc5d3
score +1.99 (about 158 million nodes searched)
11.02 ply 51:44 b6 Rb8 b7 Bd7 Ne5 Be8 Na6
score +2.73 (about 230 million nodes searched)
<I quit after 83 minutes, 350 million nodes, still searching on 11.06,
60ply q-search max>
Maybe Ed could give us solution times for Rebel 10/Anand edition in this
postion?
Hiarcs 6 (DOS)
Bxa8 fails low at 8 ply after 170", score 219 (1 pawn=128)
after 10:47 (8ply): b6 score 279
after 15:44 (8ply): g4 fail high confirmed, score 394
A good result, especially at this search depth.
Shredder 2 is a bit too slow in this position:
Shredder 2.0
computer speed: 256
searchhash: 2^20 entries per side <-> 49152 Kbytes memory
evalhash: 2^20 entires per side <-> 16384 Kbytes memory
10 -> 2:48.56 +5.07 Bxa8 Ng3+ hxg3 hxg3 Rxf8+ Qxf8 Bf3 Qh6+ Bh5 Qg5
Qg4 Bxg4 Nxa4 (8.455.502)
11.01 3:28.45 +4.82-- Bxa8 (10.531.304)
11.01 3:56.71 +4.32-- Bxa8 (12.047.681)
11.01 7:27.70 +1.81 Bxa8 Ng3+ hxg3 hxg3 Bd5+ Kh7 Rf4 Rh8 Nf2 gxf2
Rb4 Bxf4 Bg8+ Qxg8 (21.704.496)
11.02 21:47.25 +1.82++ Ne5 (61.263.796)
11.02 53:35.97 +2.09 Ne5 Ra7 Nxg6 Rff7 Nf4 Bxf4 Rxf4 Qg5
(150.564.536)
11.05 69:48.65 +2.10++ b6 (195.492.839)
11.05 76:39.48 +2.45++ b6 (214.382.568)
11.05 100:22.18 +2.72 b6 Rb8 b7 Bd7 Na6 Kg7 Nf4 Nd6 (279.154.409)
11.07 113:44.63 +2.73++ g4 (312.733.622)
11.07 119:43.37 +3.08++ g4 (328.017.956)
CSTal Paris version (zip archive dated April 1998):
Of special interest when judging the performance of Chess System Tal
might be the fact that quiescence search (e.g. Fritz peaked out at about
36 ply in the q-search at 12 ply main search depth) and king safety play
important roles here... So I don't quite agree with the following
statement (left anonymous to protect the guilty):
> However, Hsu's GMs wouldn't have needed CSTal to tell him
> that what was lacking in the DB program, as in any
> materialistic-Hyatt style program, was dynamic stuff to
> deal with that which the quiescent-seeking-Hyattian-nonsense
> tried to ignore. Namely positions leading to mating attacks. Duh.
BTW: Crafty plays b6 after a couple of seconds, just check Bob's post on
this ... :-)
Here's the CSTal logfile, all times are in minutes (meaning it needed
about 80 minutes to play 26.b6 and slightly more than 2 hours to play
26.g4):
000:38 3492n/s d6 +6.95 e4a8 f5g3 h2g3 h4g3 f1f8 d8f8 a8d5 g8h8 d5f3
f8h6 f3h5 h6h5 NULL
003:51 3630n/s d7 +6.52 e4a8 f5g3 h2g3 h4g3 f1f8 d8f8 a8d5 g8h7 d5f3
f8h6 NULL
009:12 3576n/s d8 +6.52 e4a8 f5g3 h2g3 h4g3 f1f8 d8f8 NULL
041:14 3554n/s d9 ! e4a8 f5g3 NULL
041:30 3559n/s d9 +1.41 e4a8 f5g3 h2g3 h4g3 f1f8 d8f8 a8d5 g8h8 NULL
080:46 3578n/s d9 +4.77 b5b6 a8b8 b6b7 c8d7 d3e5 d7e8 d1g4 NULL
128:42 3500n/s d9 +5.52 g2g4 h4g3 e4a8 d8h4 b1b2 a4a3 NULL
This page took 0.01 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.