Author: Tord Romstad
Date: 07:27:30 03/01/06
Go up one level in this thread
On March 01, 2006 at 09:49:44, Bernhard Bauer wrote: >Here is a comparison for a position from Evans,L - Pilnick,C 1947 > >[D]2r2rk1/p3bb1p/2n1Q1p1/q2pP3/3P1P2/p1NB1NR1/1P4P1/1K1R4 w > >Glaurung 1.02 looks at less nodes, runs a bit faster and takes less time. That Glaurung 1.0.2 is a tiny bit faster (measured in nodes/second) is normal. That Glaurung 1.0.2 needs slightly fewer nodes is just accidentally true for this particular position, I think. >However, Glaurung SMP gives some hash information under arena. :-) Yes, that's one of the few visible differences. :-) >Position from Evans,L - Pilnick,C 1947 > >FEN: 2r2rk1/p3bb1p/2n1Q1p1/q2pP3/3P1P2/p1NB1NR1/1P4P1/1K1R4 w - - 0 1 > >Glaurung-SMP: > 2/9 00:00 358 0 -0,92 Qe6h3 a3xb2 Kb1xb2 > 3/13 00:00 816 54.400 -1,23 Qe6h3 a3xb2 Nc3a2 > 4/13 00:00 2.301 153.400 -1,59 Qe6h3 a3xb2 Nc3a2 Nc6b4 Na2xb4 Be7xb4 > 5/18 00:00 10.322 219.617 -3,73 Qe6h3 a3xb2 Rd1h1 Qa5a1+ Kb1c2 Nc6b4+ > 6/24 00:00 48.860 223.105 -4,70 Qe6h3 a3xb2 Bd3f5 Qa5a1+ Kb1c2 Nc6b4+ > 6/24 00:00 90.011 240.029 -4,54 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 7/25 00:00 161.808 246.658 -4,37 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 8/27 00:01 352.101 247.609 -4,76 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 9/33 00:05 1.251.998 247.332 -5,03 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 9/33 00:06 1.647.188 246.880 -4,96 Qe6h3 a3xb2 Bd3f5 Qa5a1+ Kb1c2 Nc6b4+ > 9/33 00:09 2.327.760 251.649 -4,28 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 10/33 00:11 2.898.106 252.690 -3,26 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 11/37 00:35 8.992.061 256.461 0,00 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 12/44 00:49 12.779.558 260.143 -0,43 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 13/45 02:55 46.376.212 264.817 +1,25 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > >FEN: 2r2rk1/p3bb1p/2n1Q1p1/q2pP3/3P1P2/p1NB1NR1/1P4P1/1K1R4 w - - 0 1 > >Glaurung: > 2/9 00:00 360 1.153 -0,89 Qe6h3 a3xb2 Kb1xb2 > 3/13 00:00 819 2.625 -1,21 Qe6h3 a3xb2 Nc3a2 > 4/13 00:00 2.303 7.381 -1,56 Qe6h3 a3xb2 Nc3a2 Nc6b4 Na2xb4 Be7xb4 > 5/24 00:00 10.584 29.481 -3,70 Qe6h3 a3xb2 Rd1h1 Qa5a1+ Kb1c2 Nc6b4+ > 6/24 00:00 46.625 90.358 -4,67 Qe6h3 a3xb2 Bd3f5 Qa5a1+ Kb1c2 Nc6b4+ > 6/24 00:00 88.658 131.931 -4,42 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 7/25 00:01 159.849 170.596 -4,31 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 8/29 00:01 351.819 206.587 -4,64 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 9/33 00:05 1.217.369 214.627 -4,96 Qe6d7 a3xb2 e5e6 Qa5a1+ Kb1c2 Nc6xd4+ > 9/35 00:09 2.245.102 232.123 -4,25 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 10/35 00:12 2.827.631 238.739 -3,23 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 11/38 00:29 7.577.175 256.583 +0,01 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 12/44 00:43 11.050.487 257.833 -0,42 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 > 13/46 02:15 36.734.314 271.886 +1,26 Bd3xg6 Bf7xe6 Bg6xh7+ Kg8h8 Rd1h1 Be7h4 As a comparison, here is the analysis of Glaurung CCT8 using 2 CPUs (more specifically, an Intel Core Duo 2 GHz). The time in centiseconds is listed in the third column of each PV line: 3 -200 30 1007 Qh3 axb2 Na2 4 -250 31 3388 Qh3 axb2 Na2 Nb4 Nxb4 Bxb4 5 -507 34 14279 Qh3 axb2 Rh1 Qa1+ Kc2 Nb4+ Kd2 b1=N+ Nxb1 Qa2+ Ke3 Nxd3 Kxd3 5 -457 38 33030 Bf5 Bxe6 Bxe6+ Kg7 Bxc8 Rxc8 6 -587 40 42694 Bf5 Bxe6 Bxe6+ Kg7 Bxc8 Rxc8 Rc1 axb2 Kxb2 6 -543 48 84431 Qd7 axb2 Bf5 Rc7 Qxc7 Qxc7 Kxb2 7 -543 60 152861 Qd7 axb2 Bf5 Rc7 Qxc7 Qxc7 Kxb2 8 -575 127 524105 Qd7 axb2 e6 Qa1+ Kc2 Nxd4+ Nxd4 Rxc3+ Kd2 Rxd3+ Rxd3 Qxd1+ Kxd1 b1=Q+ Ke2 Qa2+ Rd2 Qc4+ Rd3 9 -639 331 1657462 Qd7 axb2 e6 Kh8 exf7 Qa1+ Kc2 Nxd4+ Nxd4 Rxc3+ Kd2 Rc1 Qxe7 Rxd1+ Ke3 Rxd3+ Kf2 9 -571 455 2371865 Qh3 axb2 Bf5 Qa1+ Kc2 Nb4+ Kd2 Rxc3 Bb1 Rfc8 Ke1 Qa6 9 -418 548 2922810 Bxg6 Bxe6 Bxh7+ Kh8 Bd3 Bh4 Nxh4 axb2 Ng6+ Kg7 Nxf8+ Kxf8 10 -351 683 3706890 Bxg6 Bxe6 Bxh7+ Kh8 Bd3 Bh4 Nxh4 a2+ Nxa2 Nxd4 Ng6+ Kg7 Ne7+ Kf7 Nxc8 Rxc8 11 -184 1364 7673622 Bxg6 Bxe6 Bxh7+ Kh8 Bd3 Bh4 Rh1 a2+ Nxa2 Rxf4 Nxh4 Rf1+ Rxf1 Nxd4 Ng6+ Kg7 12 +50 3288 18985574 Bxg6 Bxe6 Bxh7+ Kh8 Bd3 Bh4 Rh1 a2+ Nxa2 Rxf4 Nxh4 Rf1+ Bxf1 Qe1+ Nc1 Qe4+ Ka1 Bg4 Nf3+ Kg7 13 +90 6059 35688150 Bxg6 Bxe6 Bxh7+ Kh8 Bd3 Bh4 Rh1 Qd8 Rxh4+ Qxh4 Nxh4 Ne7 Rg6 Nxg6 Nxg6+ Kg7 Nxf8 Rxf8 14 +306 21994 132461420 Bxg6 Bxe6 Bxh7+ Kh8 Bd3 Bh4 Rh1 Rxf4 Nxh4 Rf1+ Bxf1 Rc7 Nf3+ Rh7 Rxh7+ Kxh7 Ng5+ Kh6 Nxe6 axb2 Kxb2 Qb4+ Ka1 Nxd4 Nxd5 Glaurung is much faster on 2 CPUs, which is of course not unexpected. A rather strange difference, however, is that the scores differ a lot, even when the PVs are identical. For instance, we both get the line Qh3 axb2 Na2 at the third iteration with Glaurung CCT8, but in your case the score is -1.23, while in my case it is -2.00. This has nothing to do with the fact that you used one thread while I used two; I still get -2.00 when I reduce the number of threads to 1 on my machine. Did you use parameter settings different from the defaults? If not, I suspect that there is a bug in your GUI. The most likely bug is that the GUI does not set the UCI_AnalyseMode parameter to "true" before entering analysis mode. For reference, here is Glaurung's analysis with a single thread from the above position on my machine. As you can see, the parallel search speedup is quite good in this position: 2 -1220 0 363 b4 Qxb4+ Ka1 Bxe6 2 -179 0 376 Qh3 axb2 3 -200 0 847 Qh3 axb2 Na2 4 -250 1 2320 Qh3 axb2 Na2 Nb4 Nxb4 Bxb4 5 -507 4 11318 Qh3 axb2 Rh1 Qa1+ Kc2 Nb4+ Kd2 b1=N+ Nxb1 Qa2+ Ke3 Nxd3 Kxd3 5 -457 11 31323 Bf5 Bxe6 Bxe6+ Kg7 Bxc8 Rxc8 6 -587 13 38329 Bf5 Bxe6 Bxe6+ Kg7 Bxc8 Rxc8 Rc1 axb2 Kxb2 6 -543 27 80975 Qd7 axb2 Bf5 Rc7 Qxc7 Qxc7 Kxb2 7 -543 48 149204 Qd7 axb2 Bf5 Rc7 Qxc7 Qxc7 Kxb2 8 -425 185 574129 Qd7 Rc7 Bxg6 Qb4 Bxf7+ Kh8 Rd2 Rxd7 Bxd5 Rc8 9 -646 439 1392696 Qd7 axb2 e6 Qa1+ Kc2 Nxd4+ Kd2 Rxc3 exf7+ Rxf7 Qe8+ Rf8 Rxg6+ Kh8 Rxa1 bxa1=Q Qxe7 Rxf4 9 -571 609 1925833 Qh3 axb2 Bf5 Qa1+ Kc2 Nb4+ Kd2 Rxc3 Bb1 Rfc8 Ke1 Qa6 9 -471 855 2721225 Bxg6 Bxe6 Bxh7+ Kh8 Rh1 Bh4 Rxh4 Rxf4 Rxf4 axb2 Rf6 Qxc3 Rxe6 Kxh7 10 -267 1159 3673243 Bxg6 Bxe6 Bxh7+ Kh8 Rh1 Bh4 Nxh4 Nxe5 Bf5 Ng4 Nf3+ Kg7 Rxg4+ Kf6 11 -42 2603 8185660 Bxg6 Bxe6 Bxh7+ Kh8 Rh1 Bh4 Rxh4 Rxf4 Rh6 Rg4 Bf5+ Kg7 Rh7+ Kf8 Rh8+ Kf7 Rh7+ Ke8 Bxg4 Bxg4 Rxg4 axb2 Kxb2 12 +65 6239 19532163 Bxg6 Bxe6 Bxh7+ Kh8 Rh1 Bh4 Bd3 Qd8 Rxh4+ Qxh4 Nxh4 Ne7 Ng6+ Nxg6 Rxg6 axb2 Rxe6 Rxc3 13 +90 11062 35278925 Bxg6 Bxe6 Bxh7+ Kh8 Rh1 Bh4 Bd3 Qd8 Rxh4+ Qxh4 Nxh4 Ne7 Rg6 Nxg6 Nxg6+ Kg7 Nxf8 Rxf8 14 +368 39756 128259064 Bxg6 Bxe6 Bxh7+ Kh8 Rh1 Rxf4 Bd3+ Bh4 Nxh4 a2+ Ka1 Rf1+ Bxf1 Rc7 Nf3+ Rh7 Rxh7+ Kxh7 Ng5+ Kg8 Nxe6+ Kf7 Ng5+ Ke7 The scores and PVs don't match exactly for the higher iterations with 1 vs 2 threads, but this is normal. Tord
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.