Author: Robert Hyatt
Date: 13:44:28 01/16/03
Go up one level in this thread
On January 16, 2003 at 15:37:07, David Rasmussen wrote:
>[D]8/4n3/1k6/1p4p1/1P4P1/1K6/3B4/8 w - - 0 1
Here is Crafty on my dual 2.8:
without egtbs:
16-> 3.50 2.07 1. Bxg5 Nd5 2. Bd2 Kc6 3. g5 Ne7 4.
Kc3 Kd5 5. Kd3 Ke5 6. Ke3 Nc6 7. Ke2
Nd4+ 8. Kf2 Kf5 9. Ke3
17 6.96 1.99 1. Bxg5 Nd5 2. Bd2 Kc6 3. g5 Ne7 4.
Bc3 Kd5 5. Bf6 Ng6 6. Kc3 Ke4 7. Kd2
Nf4 8. Bc3 Ng6 9. Bf6
17-> 8.13 1.99 1. Bxg5 Nd5 2. Bd2 Kc6 3. g5 Ne7 4.
Bc3 Kd5 5. Bf6 Ng6 6. Kc3 Ke4 7. Kd2
Nf4 8. Bc3 Ng6 9. Bf6
18 16.45 1.98 1. Bxg5 Nd5 2. Bd2 Kc6 3. g5 Ne7 4.
Bc3 Kd5 5. Bf6 Ng6 6. Kc3 Ke4 7. Kd2
Nf4 8. Bc3 Ng6 9. Ke2 Kf4
18-> 19.09 1.98 1. Bxg5 Nd5 2. Bd2 Kc6 3. g5 Ne7 4.
Bc3 Kd5 5. Bf6 Ng6 6. Kc3 Ke4 7. Kd2
Nf4 8. Bc3 Ng6 9. Ke2 Kf4
time=30.06 cpu=389% mat=0 n=86028737 fh=89% nps=2861k
ext-> chk=1957612 cap=56877 pp=547041 1rep=12954 mate=35
predicted=0 nodes=86028737 evals=34142238
endgame tablebase-> probes done=0 successful=0
SMP-> split=32284 stop=2193 data=20/64 cpu=1:57 elap=30.06
with egtbs:
16-> 7.66 2.10 1. Bxg5 Ng6 2. Kc3 Ne5 3. Bd8+ Kc6
4. g5 Kd5 5. Bf6 Ng6 6. Kd3 Nf4+ 7.
Ke3 Ng6 8. Bd4 Kc4 9. Bc5 Kc3 (s=2)
17 23.27 1.99 1. Bxg5 Nd5 2. Bd2 Kc6 3. g5 Ne7 4.
Bc3 Kd5 5. Bf6 Ng6 6. Kc2 Nf4 7. Bc3
Ke4 8. Kd2 Ng6 9. Bf6
time=30.12 cpu=365% mat=0 n=20076120 fh=89% nps=666k
ext-> chk=437628 cap=12887 pp=85989 1rep=1930 mate=4
predicted=0 nodes=20076120 evals=7750412
endgame tablebase-> probes done=37304 successful=37304
SMP-> split=12766 stop=979 data=23/64 cpu=1:50 elap=30.12
So in that position, I go from 2.8M nps down to 666K. 4:1 slowdown. That isn't
that
uncommon, although for positions it is worse than for games, since the next
search will
have a lot of cached stuff to use.
>
>In a position such as this, this is Chezzz _without_ EGTBs:
>
>I'm thinking...
>time nodes depth score pv
>00:00:00.19 146kn 09/12/14 +1.125 1. Bxg5 Ng6 2. Bf6 Kc6
> 3. Kc3 Kd5 4. Kd2 Ke4
> 5. Ke2 (146312)
>00:00:00.25 191kn 09/14/15 ---------------------------------------------------
>00:00:00.41 296kn 10/14/15 +1.085 1. Bxg5 Ng6 2. Bf6 Kc6
> 3. Kc3 Kd5 4. Kd3 Ke6
> 5. Bd4 Ne5+ 6. Bxe5 Kxe5 (296663)
>00:00:00.55 391kn 10/15/16 ---------------------------------------------------
>00:00:00.67 491kn 11/16/17 +1.085 1. Bxg5 Ng6 2. Bf6 Kc6
> 3. Kc3 Kd5 4. Kd3 Ke6
> 5. Bd4 Ne5+ 6. Bxe5 Kxe5 (491020)
>00:00:00.91 677kn 11/17/18 ---------------------------------------------------
>00:00:01.53 1150kn 12/17/19 +1.130 1. Bxg5 Ng6 2. Kc3 Ne5
> 3. Bd8+ Kc6 4. g5 Kd5
> 5. Bc7 Ng6 6. Kd3 Ke6
> 7. Ke4 (1150376)
>00:00:01.83 1393kn 12/18/19 ---------------------------------------------------
>00:00:03.63 2787kn 13/20/21 +1.075 1. Bxg5 Nd5 2. Bd2 Kc6
> 3. Kc2 Kd6 4. Kd3 Ke5
> 5. g5 Ne7 6. Bc3+ Kd5
> 7. Bd4 Ng6 (2787724)
>00:00:04.49 3467kn 13/20/23 ---------------------------------------------------
>00:00:06.83 5239kn 14/20/22 +1.040 1. Bxg5 Nd5 2. Bd2 Kc6
> 3. Kc2 Kd6 4. Kd3 Ke5
> 5. g5 Kf5 6. Kd4 Nc7
> 7. Kc3 Ke4 (5239484)
>00:00:07.85 6053kn 14/22/23 ---------------------------------------------------
>00:00:15.24 11Mn 15/22/24 +0.855 1. Bxg5 Nd5 2. Bd2 Kc6
> 3. Ka3 Kd6 4. Bg5 Kd7
> 5. Kb3 Ke6 6. Bd8 Ke5
> 7. g5 Kf5 8. Ka3 (11513787)
>00:00:18.97 13Mn 15/23/25 ---------------------------------------------------
>00:00:28.80 20Mn 16/24/27 1/11 1. Bxg5
>
> Time used: 00:00:30.00
> nps: 718426 nodes: 21559976 Q-nodes: 707976 (3%)
> move order: 94%
> hash probes: 8770930 hash hits: 5045831 (57%) hash usage: 99%
> pawn hash probes: 9470211 pawn hash hits: 9469820 (99%)
> EGTB probes: 0
>1. Bxg5
>
>This is Chezzz _with_ EGTBs:
>
>time nodes depth score pv
>00:00:00.22 139kn 09/12/14 +1.125 1. Bxg5 Ng6 2. Bf6 Kc6
> 3. Kc3 Kd5 4. Kd2 Ke4
> 5. Ke2 (139961)
>00:00:00.29 184kn 09/14/15 ---------------------------------------------------
>00:00:00.45 284kn 10/14/15 +1.085 1. Bxg5 Ng6 2. Bf6 Kc6
> 3. Kc3 Kd5 4. Kd3 Ke6
> 5. Bd4 Ne5+ 6. Bxe5 Kxe5 (284263)
>00:00:00.58 370kn 10/15/17 ---------------------------------------------------
>00:00:00.73 472kn 11/15/16 +1.085 1. Bxg5 Ng6 2. Bf6 Kc6
> 3. Kc3 Kd5 4. Kd3 Ke6
> 5. Bd4 Ne5+ 6. Bxe5 Kxe5 (472265)
>00:00:01.00 650kn 11/17/19 ---------------------------------------------------
>00:00:01.42 920kn 12/18/19 +1.130 1. Bxg5 Ng6 2. Kc3 Ne5
> 3. Bd8+ Kc6 4. g5 Kd5
> 5. Bc7 Ng6 6. Kd3 Ke6
> 7. Ke4 (920316)
>00:00:01.75 1149kn 12/18/19 ---------------------------------------------------
>00:00:03.48 2310kn 13/20/21 +1.075 1. Bxg5 Nd5 2. Bd2 Kc6
> 3. Kc2 Kd6 4. Kd3 Ke5
> 5. g5 Ne7 6. Be3 Ng6
> 7. Bd4+ Kd5 (2310354)
>00:00:04.45 2971kn 13/20/21 ---------------------------------------------------
>00:00:12.04 4334kn 14/20/23 +1.040 1. Bxg5 Nd5 2. Bd2 Kc6
> 3. Kc2 Kd6 4. Kd3 Ke5
> 5. g5 Kf5 6. Kd4 Nc7
> 7. Kc3 Ke4 (4334933)
>00:00:17.61 5058kn 14/21/23 ---------------------------------------------------
>00:00:22.70 6143kn 15/20/22 1/11 1. Bxg5
>
> Time used: 00:00:30.00
> nps: 234538 nodes: 7038510 Q-nodes: 196510 (2%)
> move order: 95%
> hash probes: 3162813 hash hits: 2220367 (70%) hash usage: 87%
> pawn hash probes: 2866265 pawn hash hits: 2865984 (99%)
> EGTB probes: 25055
>1. Bxg5
>
>That is quite a speed drop. What I don't understand is that during this search,
>the disk is working like crazy. I only probe in the first 2 plys. There is 11
>legal moves in the position, so that's roughly 100 positions. That can't take
>that much activity to probe. And then I only probe within the iteration depth,
>and only then if there have been a capture or a promotion, and only if the
>number of pieces on the board is small enough (<= 5 for 3-5 man tables). In this
>position, there is of course a capture now and then, and a promotion sometimes
>too. But why this constant disk access? It really feels to me as if it's the
>disk I/O that's killing me, and not because I have a slow disk, but because
>there is far too much disk activity for some reason. And the reason isn't that I
>probe too often due to a bug, because my probe numbers are like everybody elses.
>
>You can see that I get about 25000 probes on a 30 second think on an Athlon XP
>1800+. How much do you get?
>
>/David
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.