Author: Uri Blass
Date: 14:41:18 08/10/02
Go up one level in this thread
On August 10, 2002 at 16:47:05, Michael Babigian wrote:
>{snip}
>>New position
>>[d]r1b1r1k1/1q3ppp/ppn5/2bNp3/P4B2/5Q1P/BP3PP1/R2R2K1 w - - 0 1
>
>{snip}
>
>>Yes it is very good. For some rough comparisons, DB Jr had 16 chess processors
>>at 2M nodes per second each. Roughly 32M nodes per second. But then Hsu
>>claimed about 30% efficiency so make that 9M nodes per second. They took
>>about 2 minutes to find the +2.5 score (I am trying to locate this old email
>>to get the exact output and time but I only have about a million emails I
>>have sent over the past 7+ years so finding that one is non-trivial).
>>
>>In any case, you can probably compare them at 9M nodes per second times 120
>>seconds, vs Shredder at whatever time it took...
>>
>>One thing is to see if Shredder has found _the_ line yet. IE the line played
>>in the game was optimal. DB Jr had every move right for about 10 moves until
>>the PV ran out due to the hardware... so it saw what was really going to happen
>>pretty well...
>
>It took me a while to get around to examining this position, but I think it does
>illustrate how far ahead DB Jr was in 1996. A gave Shredder 6.02 a good 21
>hours on my dual Athlon 1800+ (768MB Hash) and it produced the following
>analysis.
>
> 8.01 0:01 -0.59 19.Be3 Be6 20.Rac1 Bxe3 21.Qxe3 Rad8 22.b4 f5 23.Bc4
>(561.326) 528.5
> 9.01 0:02 -0.59 19.Be3 Be6 20.b4 Bxe3 21.Qxe3 Rad8 22.Rac1 f5 23.b5 axb5
>(1.497.753) 518.2
> 9.03 0:03 -0.58++ 19.Bh6 Re6 20.Bg5 h6 21.Bh4 e4 22.Qg3 (1.768.791) 514.6
> 9.03 0:03 -0.45 19.Bh6 Re6 20.Be3 Nd4 21.Qh5 g6 22.Qh4 Ne2+ 23.Kh2 Nd4
>(1.944.656) 512.1
> 10.01 0:05 -0.51 19.Bh6 Re6 20.Be3 Nd4 21.Qh5 Bd7 22.b4 g6 23.Qh4 Ne2+
>24.Kf1 Bxe3 25.fxe3 (2.917.469) 513.0
> 10.02 0:06 -0.50++ 19.Be3 Be6 20.b4 Bxe3 21.Qxe3 Rad8 22.Qg3 Bxd5 23.Bxd5 b5
>24.axb5 Nd4 25.Bxb7 axb5 (3.181.395) 510.3
> 10.02 0:07 -0.42 19.Be3 Be6 20.Rac1 Bxe3 21.Qxe3 Rac8 22.Qg3 Kh8 23.b4 f5
>24.Bc4 (3.858.859) 508.1
> 11.01 0:16 -0.42 19.Be3 Be6 20.Rac1 Bxe3 21.Qxe3 b5 22.Qg3 Kh8 23.axb5 axb5
>24.Bb3 f5 25.Qe3 (8.086.388) 501.0
> 11.03 0:19 -0.41++ 19.Bh6 Re6 20.Be3 Nd4 21.Qh5 Bd7 22.b4 g6 23.Qg5 Ne2+
>24.Kf1 Nf4 25.bxc5 Nxd5 (9.726.667) 499.2
> 11.03 0:20 -0.41 19.Bh6 Re6 20.Be3 Nd4 21.Qh5 Bd7 22.b4 Bd6 23.Bb1 g6
>24.Ne7+ Bxe7 25.Bxd4 (10.099.781) 499.1
> 11.06 0:23 -0.40++ 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Qxf6 e4 23.Qxf7+ Kh8
>24.Bxf8 (12.105.226) 507.0
> 11.06 0:26 -0.05++ 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Qxf6 e4 23.Qxf7+ Kh8
>24.Bxe4 Nd4 25.Bxb7 Re1+ 26.Rxe1 (13.273.162) 509.8
> 11.06 0:41 +0.51 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Qxg7+ Kxg7 26.Bxb7 Bc2 (21.064.708) 513.7
> 12.01 0:54 +0.67 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Rg3 Bg6 26.Qxg7+ Kxg7 27.Bxb7 Ra7 28.Rd7 (28.398.743) 517.5
> 13.01 1:37 +0.73 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Rg3 Bg6 26.Qxg7+ Kxg7 27.Bxb7 (50.819.633) 520.3
> 14.01 2:43 +0.98++ 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Bxb7 Bxf6 26.Bxa8 Bxh3 27.gxh3 Rxa8 (85.465.818) 521.1
> 14.01 3:50 +1.48++ 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Bxb7 Bxf6 26.Bxa8 Kh8 27.Bd5 Bc2 28.Bxf7 (120.728.370) 523.7
> 14.01 5:10 +2.04 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Bxb7 Bxf6 26.Bxa8 Rxa8 27.Rf3 Bc2 28.Rc1 Bxa4 (163.099.548) 525.4
> 15.01 9:20 +2.00 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Bxb7 Bxf6 26.Bxa8 Rxa8 27.Rf3 Bxh3 28.Rxf6 Be6 29.Re1
>(297.246.494) 530.5
> 16.01 23:35 +1.98 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Ra3 Bxh6 23.Qxf6 Bf5
>24.Bxc6 Bg7 25.Bxb7 Bxf6 26.Bxa8 Rxa8 27.Rf3 Bxh3 28.Rxf6 Be6 29.Rh6
>(750.474.835) 530.1
> 17.01 50:05 +2.23++ 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bd7 22.Bxf8 Kxf8 23.Qxf6
>Re6 24.Bxe6 Bxe6 25.Rd6 Kg8 26.Rad1 Rf8 27.R1d3 Ne7 28.Rd7 Bxd7 29.Qxe7 f6
>30.Rxd7 (1.596.353.875) 531.1
> 17.01 69:48 +2.31 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8
>24.Bc6 f5 25.Bd5 Bf4 26.Re1 Be6 27.Rad1 Rd8 28.g3 Rxd5 29.Rxd5 (2.206.598.336)
>526.7
> 18.01 146:48 +2.39 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bxh6 22.Bxc6 Qe7 23.Bxe8
>Rb8 24.Bc6 f5 25.Bd5 Bf4 26.Re1 Qf6 27.Rad1 Bd7 28.Qa3 Rd8 (4.656.973.467) 528.7
> 19.01 374:30 +2.43 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bxh6 22.Bxc6 Qe7 23.Bxe8
>Rb8 24.Bc6 f5 25.Ra3 Bg7 26.Qd3 Kf8 27.Qe2 (11.909.898.262) 530.0
> 20.01 1230:57 +2.61 19.Nf6+ gxf6 20.Bh6 Bf8 21.Bd5 Bxh6 22.Bxc6 Qe7 23.Bxe8
>Rb8 24.Qh5 Kg7 25.Ra3 Bf4 26.Bc6 Be6 27.Be4 Rd8 28.Qxh7+ Kf8 (38.781.207.562)
>525.0
>
>As you can see at depth 16 it believes 23 ...Bf5 is a better defense for black
>and later replaces 22. Ra3 with 22. Bxf8. Finally at depth 17 and above it
>settles on 21 ... Bxh6 as blacks best defense in this lost position.
>
>Your comments that the game line was optimal got me curious why Shredder came to
>a different conclusion. At your estimation of 9M nodes/second, Shredder would
>take about 36 minutes to do the work DB Jr did in 2. Granted the shape of the
>search tree is radically different and could account for the difference in the
>line, I continued the experiment further.
>
>I moved Shredder forward to analyze at blacks move 21. I gave it 3.5 hours at
>this move and it produced the following.
>
> 8.01 0:00 -0.06 21...Bd7 22.Qxf6 Re6 23.Bxe6 Bxe6 24.Qg5+ Kh8 25.Rd8 Nxd8
>26.Qxe5+ f6 27.Qxf6+ Bg7 28.Qf8+ Bxf8 29.Bg7+ Qxg7 (535.380) 571.3
> 9.01 0:01 -0.06 21...Bd7 22.Qxf6 Re6 23.Bxe6 Bxe6 24.Qg5+ Kh8 25.Rd8 Rxd8
>26.Qf6+ Kg8 27.Qg5+ Kh8 28.Qf6+ Kg8 29.Qg5+ Kh8 30.Qf6+ (1.100.522) 577.3
> 10.01 0:04 -0.07 21...Bd7 22.Qxf6 Re6 23.Bxe6 Bxe6 24.Qg5+ Kh8 25.Bxf8 Rxf8
>26.Rd7 Qxd7 27.Qf6+ Kg8 28.Qg5+ Kh8 29.Qf6+ Kg8 30.Qg5+ Kh8 31.Qf6+ (2.800.298)
>580.0
> 11.01 0:12 -0.07 21...Bd7 22.Qxf6 Re6 23.Bxe6 Bxe6 24.Qg5+ Kh8 25.Bxf8 Rxf8
>26.Rd7 Qxd7 27.Qf6+ Kg8 28.Qg5+ Kh8 29.Qf6+ Kg8 30.Qg5+ Kh8 31.Qf6+ (7.159.038)
>574.8
> 12.01 0:31 -0.07 21...Bd7 22.Qxf6 Re6 23.Bxe6 Bxe6 24.Qg5+ Kh8 25.Bxf8 Rxf8
>26.Rd7 Qxd7 27.Qf6+ Kg8 28.Qg5+ Kh8 29.Qf6+ Kg8 30.Qg5+ Kh8 31.Qf6+ (18.086.566)
>571.6
> 13.01 1:28 -0.32-- 21...Bd7 22.Bxf8 Rxf8 23.Qxf6 e4 24.Ra3 e3 25.Rxe3 Rfe8
>26.Bxf7+ Kf8 27.Bxe8+ Kg8 28.Rxd7 (49.373.205) 557.2
> 13.01 1:44 -0.82-- 21...Bd7 22.Bxf8 Rxf8 23.Qxf6 e4 24.Ra3 e3 25.Rxe3 Bxh3
>26.gxh3 a5 (58.113.016) 554.9
> 13.01 5:12 -2.98 21...Bd7 22.Bxf8 Kxf8 23.Qxf6 Re6 24.Bxe6 Bxe6 25.Rd6 Kg8
>26.Rad1 Qc7 27.Rxe6 fxe6 28.Qxe6+ Qf7 29.Qxc6 Rb8 30.Rd7 Qf4 31.Qc2
>(174.047.507) 556.3
> 13.02 5:20 -2.97++ 21...e4 22.Qg3+ Kh8 23.Bxf8 Rxf8 24.Qd6 Qe7 25.Qxe7 Nxe7
>26.Bxa8 Bxh3 27.Bxe4 Bf5 28.Bxf5 Nxf5 29.Rac1 (178.766.713) 557.2
> 13.02 6:02 -2.84 21...e4 22.Qg3+ Kh8 23.Bxf8 Rxf8 24.Qd6 Qe7 25.Qxe7 Nxe7
>26.Bxa8 Bxh3 27.Bxe4 Be6 28.Bb7 Bc8 29.Bxc8 Rxc8 30.Rd7 Re8 31.Rad1
>(202.552.782) 558.9
> 13.03 6:09 -2.83++ 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Qh5 Bg5 25.f4
>(206.577.969) 558.4
> 13.03 6:12 -2.48++ 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Qh5 Bg5 25.f4 Qxe8
>26.fxg5 fxg5 27.Rf1 Kh8 28.Qxf7 (207.847.700) 558.5
> 13.03 6:22 -2.08 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 f5 25.Qd3 Bg5
>26.Bd5 Qf8 27.Re1 Qe8 28.Qg3 (213.440.345) 557.4
> 13.15 6:50 -2.07++ 21...Nd4 22.Qg3+ Kh8 23.Bxb7 Ne2+ 24.Kf1 Nxg3+ 25.Ke1
>(228.488.896) 556.5
> 13.15 7:10 -2.07 21...Nd4 22.Rxd4 Qe7 23.Bxf8 Qxf8 24.Rd3 e4 25.Bxe4 Ra7
>26.Bc6 Rd8 27.Rxd8 Kh8 28.Rxf8+ Kg7 29.Rxc8 (240.020.215) 557.2
> 14.01 7:44 -2.32-- 21...Nd4 22.Rxd4 Qe7 23.Bxf8 Kxf8 24.Rh4 Ra7 25.Rxh7 Be6
>26.Be4 f5 27.Bxf5 Qf6 28.Qa3+ Kg8 29.Qg3+ Qg7 30.Rxg7+ (258.088.329) 556.1
> 14.01 8:41 -2.82-- 21...Nd4 22.Rxd4 Qe7 23.Bxf8 Kxf8 24.Rh4 Ra7 25.Rxh7 Be6
>26.Be4 f5 27.Qh5 Qf6 28.Rh6 Qd8 29.Bxf5 Kg8 30.Bxe6 Rc7 31.Qxe5 (289.464.148)
>554.9
> 14.01 11:46 -3.41 21...Nd4 22.Rxd4 Qe7 23.Bxf8 Qxf8 24.Rd3 e4 25.Bxe4 Ra7
>26.Qg3+ Kh8 27.Qh4 h6 28.Rg3 Re6 29.Bf5 Rc6 30.Bxc8 Rxc8 31.Qxf6+ Kh7
>(387.983.968) 549.4
> 14.02 13:13 -3.40++ 21...Bd7 22.Bxf8 Kxf8 23.Qxf6 Re6 24.Bxe6 Bxe6 25.Rd6 Kg8
>26.Rad1 Qc7 27.Rxe6 fxe6 28.Qxe6+ Qf7 29.Qxc6 Rb8 30.Rd7 Qf5 31.Qd5+ Kh8 32.Kh2
>(435.243.778) 548.8
> 14.02 14:09 -3.05++ 21...Bd7 22.Bxf8 Kxf8 23.Qxf6 Re6 24.Bxe6 Bxe6 25.Rd6 Kg8
>26.Rad1 Qc7 27.Rxe6 fxe6 28.Qxe6+ Qf7 29.Qxc6 Rf8 30.Qxb6 Qf4 31.b4 Kh8 32.Rd7
>(468.274.897) 551.5
> 14.02 17:42 -2.96 21...Bd7 22.Bxf8 Kxf8 23.Qxf6 Re6 24.Bxe6 Bxe6 25.Rd6 Kg8
>26.Rad1 Qc7 27.Rxe6 fxe6 28.Qxe6+ Qf7 29.Qxc6 Rf8 30.Qxb6 Qc4 31.Qe3 Qf4 32.Qxf4
>(584.465.184) 549.8
> 14.03 18:33 -2.95++ 21...e4 22.Qg3+ Kh8 23.Bxf8 Rxf8 24.Qd6 Qe7 25.Qxe7 Nxe7
>26.Bxa8 Bxh3 27.Bxe4 Be6 28.Bb7 Bc8 29.Bxc8 Nxc8 30.b4 (611.571.563) 549.2
> 14.03 19:09 -2.93 21...e4 22.Qg3+ Kh8 23.Bxf8 Rxf8 24.Qd6 Qe7 25.Qxe7 Nxe7
>26.Bxa8 Bxh3 27.Bxe4 Be6 28.Bb7 Bc8 29.Bxc8 Nxc8 30.Rd7 Kg7 (631.983.159) 549.6
> 14.04 19:15 -2.92++ 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 (635.262.043)
>549.6
> 14.04 19:21 -2.57++ 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 Kh8 25.Qh5 Bg7
>26.Be4 h6 27.Rd8+ Qxd8 28.Qd1 Qxd1+ 29.Kh2 (638.302.731) 549.7
> 14.04 19:43 -2.03 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 f5 25.Qd3 Bg5
>26.Bd5 Bf4 27.Re1 Qb4 (649.720.458) 548.9
> 15.01 21:27 -2.10 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 f5 25.Qd5 e4
>26.Ra3 Bf4 27.Qd4 Be5 28.Qd8+ Qf8 29.b4 (700.748.819) 544.1
> 16.01 33:42 -2.17 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 f5 25.Bd5 Bf4
>26.Re1 Be6 27.Rad1 b5 (1.077.021.858) 532.6
> 17.01 60:17 -2.16 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 f5 25.Bd5 Bf4
>26.Re1 Be6 27.Bxe6 Qxe6 28.Qd3 b5 29.axb5 (1.902.106.464) 525.7
> 18.01 123:10 -2.16 21...Bxh6 22.Bxc6 Qe7 23.Bxe8 Rb8 24.Bc6 f5 25.Bd5 Bg7
>26.Rac1 Be6 27.Qe2 Bc8 28.b3 Bb7 29.Qh5 (3.849.522.935) 520.8
>
>Again, Shredder eventually gives up on 21...Bd7 and settles on 21...Bxh6 as best
>for black. Here again Shredder has enough time to out search the 1 billion
>nodes DB Jr. saw in 2 minutes.
>
>Finally I tried one more experiment by selecting the Shredder Brute Force
>engine. I figured this might better approximate DB Jr's search tree shape.
>Again analyzing from blacks move 21 after 14.5 hours and 17B nodes Shredder
>stayed with Bd7 to depth 18.
>
>Clearly Shredder understands what is happening in about 5 minutes on my
>hardware, but considering how much faster Deeper Blue was than DB Jr, this
>definitely illustrates how far behind the latest PC software is compared to the
>now disassembled monster.
I disagree with you.
You cannot decide based on one position that the latest PC software is behind
Deep blue.
The choice of this position is not a random choice and you choose on purpose
position when deeper blue is good.
It is also possible to find positions when chessmaster is 100 times faster than
other programs but it does not do chessmaster better.
I think that it is better to look at the games of deeper blue against kasparov
and I did not find a single impressive move or evaluation that top programs need
hours to find.
Uri
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.