Hi there,
1. Stockfish 2.1.1 JA x64 PHQ (setting)
2. Stockfish 2.1.1 JA x64 (default)
are still running in SWCR.
1.200 games for each engine are to play!
More information to the "PHQ" setting can be found in my News / Actual page.
Have fun!
Best
Frank
SWCR: SF 2.1.1 JA x64 PHQ & SF 2.1.1 JA x64 ...
Moderators: hgm, Rebel, chrisw
-
Frank Quisinsky
- Posts: 6824
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
-
Uri Blass
- Posts: 10427
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
Re: SWCR: SF 2.1.1 JA x64 PHQ & SF 2.1.1 JA x64 ...
It seems that PHQ is slightly better.
Not enough games but results so Far in Frank's page:
PHQ (setting) :332.5 - 103.5 76.26% Perf=2949
default: 327.0 - 109.0 75.00% Perf=2932
Not enough games but results so Far in Frank's page:
PHQ (setting) :332.5 - 103.5 76.26% Perf=2949
default: 327.0 - 109.0 75.00% Perf=2932
-
Frank Quisinsky
- Posts: 6824
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: SWCR: A bit statistic for this interesting compare!
Hi Uri,
a bit more on information:
1. Draw
PHQ = 31% draw after 463 games
Default = 35% draw after 459 games
2. Lost on time:
7 of 922 games.
2x ChessTiger 2007
5x SF (3x default, 2x PHQ).
3x remis under 16 moves, all from Default
This games are replayed so far!
3. This one is very interesting:
PHQ won 33 games up to move 56 with mate
Default won 16 games up to move 56 with mate
7,13% from all played games for PHQ.
The SWCR have a new number 1 in this statistic. Better results as Spark 1.0 !!!
3,48% for default ... again a bit passiver as 1.8.0, 1.9.1 and 2.0.1. Best older SF version is 1.7.1 but 2.1.1 with PHQ setting produced here a better result as 1.7.1.
And 11 ELO more for PHQ.
4% fewer remis = 2 ELO more in Bayesian calculation!
Here the actual SWCR:
Best
Frank
a bit more on information:
1. Draw
PHQ = 31% draw after 463 games
Default = 35% draw after 459 games
2. Lost on time:
7 of 922 games.
2x ChessTiger 2007
5x SF (3x default, 2x PHQ).
3x remis under 16 moves, all from Default
This games are replayed so far!
3. This one is very interesting:
PHQ won 33 games up to move 56 with mate
Default won 16 games up to move 56 with mate
7,13% from all played games for PHQ.
The SWCR have a new number 1 in this statistic. Better results as Spark 1.0 !!!
3,48% for default ... again a bit passiver as 1.8.0, 1.9.1 and 2.0.1. Best older SF version is 1.7.1 but 2.1.1 with PHQ setting produced here a better result as 1.7.1.
And 11 ELO more for PHQ.
4% fewer remis = 2 ELO more in Bayesian calculation!
Here the actual SWCR:
Code: Select all
SWCR : 109.802 games #Time control: 40/10 "repeatedly"
Last update : June 12th, 2011 #Game average: 40 minutes, move average = 85
#Resign : OFF
LIVE 1 : SF 2.1.1 JA x64 PHQ #Ponder : ON
Cores : 08/16 #Learning : OFF
To play : 0.737 of 1.200 #Endgames : Diff. 4-pieces, 32Mb cache
#Opening book: Own PGN random 4.1 book
LIVE 2 : SF 2.1.1 JA x64 #GUI : Shredder Classic 4
Cores : 08/16 #OS : Windows XP Pro. x64 Edition
To play : 0.741 of 1.200 #Processors : Intel® Core™2Q, 4xQ9550 2,83GHz
#Cores : 1 core for each engine
Database : 108.526 games #Hash-Tables : 256Mb
Updated : June 08th, 2011 #Games : *NEW* 1.000 per engine minimal
NAME / version of engine ELO + - GAM SC OP DR
-----------------------------------------------------------------------
1 Houdini 1.5 x64 3004 16 16 1712 79% 2772 28%
2 Rybka 4 x64 Exp. 42 2970 20 19 1240 80% 2720 25%
3 Rybka 4.1 x64 Exp. 79TD v.1 2966 20 19 1200 79% 2739 26%
4 Rybka 4 x64 Exp. 61 2962 21 21 1000 78% 2744 27%
5 Rybka 4.1 x64 2960 19 19 1111 76% 2766 34%
6 Stockfish 2.1.1 JA x64 PHQ 2949 30 29 463 76% 2747 31% NEW "Little experiment"
7 Houdini 1.03a x64 2945 21 21 1000 80% 2712 30%
8 IvanHoe B47cB x64 2945 18 18 1152 73% 2774 38%
9 Rybka 4 x64 2940 17 17 1520 80% 2701 29%
10 Stockfish 2.1.1 JA x64 2938 29 29 459 75% 2748 35% NEW + 29
11 Fire 1.5 xTreme x64 2936 18 18 1232 75% 2749 34% NEW + 31
12 IvanHoe B49jA x64 2934 19 18 1160 76% 2740 34%
13 IvanHoe B52aC x64 2920 20 20 1000 76% 2728 34%
14 Stockfish 2.0.1 JA x64 2909 18 18 1120 69% 2770 38%
15 Stockfish 1.9.1 JA x64 2907 18 17 1280 73% 2729 33%
16 Stockfish 1.8.0 JA x64 2907 18 18 1200 75% 2716 33%
17 Fire 1.31 x64 2905 19 19 1040 73% 2735 37%
18 Rybka 3 x64 2904 21 20 1000 78% 2687 29%
19 Critter 1.01 x64 2899 18 17 1191 68% 2770 37%
20 Stockfish 1.7.1 JA x64 2898 19 18 1120 76% 2708 34%
21 Stockfish 1.9.1 JA w32 2894 20 20 1000 77% 2693 31%
22 Rybka 4 w32 2892 18 18 1200 76% 2696 32%
23 Critter 0.90 x64 2873 17 17 1200 68% 2743 37%
24 Stockfish 1.7.1 JA w32 2873 18 18 1200 75% 2688 31%
25 Stockfish 2.0.1 JA w32 2873 20 20 1000 76% 2680 34%
26 Critter 0.90 w32 2869 20 20 1000 76% 2680 32%
27 Stockfish 1.8.0 JA w32 2864 19 19 1000 73% 2703 38%
28 Rybka 3 w32 2859 16 16 1520 74% 2688 31%
29 Komodo 1.3 JA x64 2848 17 17 1192 61% 2771 39%
30 Naum 4.2 x64 2836 11 11 3232 64% 2734 36%
31 Critter 0.80 x64 2832 15 15 1680 67% 2702 33%
32 Stockfish 1.6.3 JA w32 2825 18 18 1080 71% 2678 36%
33 Naum 4.2 w32 2820 13 13 2320 68% 2696 37%
34 Critter 0.80 w32 2814 18 18 1040 67% 2694 37%
35 Critter 0.70 x64 2811 19 19 1000 65% 2705 38%
36 Naum 4.1 w32 2809 19 18 1000 67% 2692 35%
37 Komodo 1.2 JA x64 2806 13 13 2280 62% 2715 39%
38 Stockfish 1.6.0 JA w32 2801 18 18 1000 67% 2690 39%
39 Shredder 12 w32 2800 8 8 6072 62% 2716 36%
40 GullChess 1.2 x64 2791 17 17 1192 52% 2773 40%
41 Sjeng c't 2010 w32 2790 10 10 3952 59% 2727 37%
42 Shredder 12 x64 2788 15 15 1600 63% 2691 34%
43 GullChess 1.1 x64 2788 17 17 1160 56% 2744 37%
44 Komodo 1.0 JA x64 2788 19 18 1000 64% 2692 40%
45 Naum 4.0 w32 2784 18 18 1000 65% 2685 38%
46 Spike 1.4 Leiden w32 2783 10 10 3712 59% 2719 37%
47 Deep Fritz 12 w32 2779 14 14 1760 62% 2698 41%
48 Critter 0.70 w32 2776 18 18 1000 60% 2709 40%
49 Komodo 1.3 JA w32 2775 19 18 1000 63% 2684 36%
50 Protector 1.4.0 JA x64 2773 17 17 1192 50% 2774 39%
51 GullChess 1.0a x64 2773 15 15 1480 58% 2712 37%
52 Junior 12.5.0.3 x64 2765 16 16 1311 52% 2753 35% NEW + 46
53 Spark 1.0 x64 2762 14 14 1831 50% 2765 40%
54 Hiarcs 13.2 w32 2762 13 13 2190 54% 2733 40%
55 Komodo 1.2 JA w32 2758 17 17 1200 58% 2700 41%
56 GullChess 1.0a w32 2747 18 18 1000 57% 2699 40%
57 GullChess 1.1 w32 2745 18 18 1000 59% 2685 40%
58 Fritz 12 w32 2744 17 17 1160 59% 2687 44%
59 Spark 0.5 x64 2741 13 13 2120 55% 2705 36%
60 Hiarcs 13.1 w32 2738 11 11 2920 52% 2720 38%
61 Thinker 5.4d Inert x64 2737 11 10 3231 50% 2735 38%
62 Stockfish 1.5.1 JA w32 2727 18 18 1000 59% 2671 43%
63 Zappa Mexico II x64 2726 11 10 3230 49% 2735 39%
64 Spark 1.0 w32 2719 18 18 1000 55% 2686 36%
65 Junior 12.0 x64 2719 14 14 1840 45% 2752 32%
66 Protector 1.4.0 JA w32 2717 18 18 1000 54% 2686 36%
67 Spark 0.4 x64 2716 18 18 1000 53% 2695 40%
68 Komodo 1.0 JA w32 2714 17 17 1200 53% 2693 40%
69 Fruit 09_07_05 x64 2714 11 11 3230 47% 2735 33%
70 Protector 1.3.4 JA x64 2713 14 14 1720 50% 2712 37%
71 Thinker 5.4d Inert w32 2709 11 11 2880 52% 2698 42%
72 Spark 0.5 w32 2707 16 16 1200 51% 2702 41%
73 Protector 1.3.6-370 JA x64 2706 17 17 1280 46% 2736 33%
74 Critter 0.60 x64 2703 18 18 1000 51% 2695 39%
75 Junior 12.0 w32 2699 19 18 1000 51% 2687 35%
76 Booot 5.1.0 w32 2696 18 18 1080 42% 2756 37%
77 Protector 1.3.5 x64 2694 18 18 1000 49% 2704 41%
78 Sjeng WC-2008 x64 2688 14 14 1760 47% 2712 36%
79 Doch 1.3.4 JA w32 2687 18 18 1000 50% 2684 42%
80 Junior 11.2 w32 2686 17 17 1200 48% 2703 33%
81 Critter 0.60 w32 2685 18 18 1000 49% 2695 39%
82 Hannibal 1.0a x64 2684 12 12 2670 42% 2745 34%
83 Junior 11.2 x64 2683 15 16 1520 47% 2708 30%
84 Spark 0.4 w32 2680 18 18 1000 48% 2693 42%
85 Protector 1.3.4 JA w32 2678 14 15 1600 47% 2700 39%
86 Onno 1.2.70 x64 2678 11 11 2990 42% 2735 35%
87 Cyclone xTreme Wrath w32 2674 17 17 1080 47% 2696 41%
88 Onno 1.1.1 x64 2674 18 18 1000 47% 2697 40%
89 Protector 1.3.5 JA w32 2673 18 18 1000 45% 2705 43%
90 Protector 1.3.2 w32 2672 17 17 1160 47% 2694 41%
91 Equinox 0.95 x64 2672 18 18 1190 36% 2777 31%
92 Doch 1.2 JA w32 2669 18 18 1000 49% 2677 38%
93 Junior 2010 w32 2668 16 16 1240 47% 2690 36%
94 Protector 1.3.1b w32 2667 18 18 1000 47% 2687 42%
95 Hiarcs 12.1 w32 2666 18 18 1000 46% 2691 40%
96 Sjeng WC-2008 w32 2664 12 12 2240 45% 2699 37%
97 Equinox 0.95 w32 2663 18 18 1000 46% 2688 36%
98 Hiarcs 12.1 w32 Sharpen PV 2663 16 16 1280 45% 2697 39%
99 Zappa Mexico II w32 2658 11 11 2880 44% 2699 41%
100 Junior 11.1a x64 2655 19 19 1000 44% 2697 33%
101 Doch 09.980 JA w32 2655 18 18 1000 47% 2674 41%
102 Spark 0.3a w32 2654 17 17 1120 44% 2696 41%
103 Spark 0.3 w32 2646 18 18 1000 43% 2692 42%
104 Junior 11.1a w32 2642 18 18 1000 43% 2693 36%
105 Hannibal 1.0a w32 2640 14 14 1720 41% 2702 35%
106 Bright 0.5c w32 2633 18 18 1040 39% 2707 35%
107 Umko 1.1 x64 2633 14 14 1950 32% 2768 32%
108 Onno 1.1.1 w32 2629 14 15 1520 40% 2694 41%
109 Onno 1.2.70 w32 2628 14 14 1840 39% 2702 39%
110 Loop M1-T x64 2627 18 18 1110 34% 2745 35%
111 Loop 2007 x64 2622 12 12 2400 36% 2725 34%
112 Loop 2007 w32 2616 14 14 1840 38% 2703 36%
113 Crafty 23.4 JA x64 2616 14 14 1990 31% 2758 32%
114 Jonny 4.00 w32 2616 11 11 3070 36% 2724 29%
115 Fruit 05/11/03 w32 2615 11 11 2880 38% 2700 40%
116 Equinox 0.87t x64 2614 17 17 1280 33% 2739 31%
117 Scorpio 2.7 JA x64 2614 18 19 1188 29% 2779 28%
118 Loop 13.6 w32 2613 15 15 1520 38% 2694 39%
119 Twisted Logic 20100131x x64 2613 18 18 1120 35% 2718 32%
120 Critter 0.52b w32 2612 18 18 1040 38% 2697 37%
121 Umko 1.0 x64, no ponder 2611 17 17 1200 34% 2726 37%
122 Glaurung 2.2 JA w32 2608 18 18 1080 37% 2699 36%
123 Ktulu 9.03 w32 2607 14 14 1760 37% 2702 31%
124 BugChess2 1.9 x64 2605 17 17 1270 38% 2693 30% NEW + 42
125 Equinox 0.83 x64 2601 18 18 1160 32% 2734 32%
126 SmarThink 1.20 x64 2601 12 12 2710 33% 2729 31%
127 Crafty 23.3 JA x64 2599 17 18 1200 33% 2726 34%
128 SmarThink 1.20 w32 2593 11 11 2880 35% 2700 36%
129 Tornado 4.40 x64 2593 16 16 1550 26% 2776 28%
130 Umko 1.1 w32 2590 19 19 1000 36% 2691 34%
131 Crafty 23.4 JA w32 2583 19 19 1000 35% 2691 34%
132 Equinox 0.83 w32 2577 19 19 1000 32% 2706 33%
133 Twisted Logic 20100131x w32 2573 15 15 1600 32% 2706 30%
134 Spike 1.2 Turin w32 2572 16 16 1480 31% 2706 34%
135 Cipollino 3.25 x64 2571 19 19 1080 30% 2725 31%
136 Tornado 4.40 w32 2567 16 16 1360 33% 2687 34%
137 Crafty 23.3 JA w32 2565 19 19 1000 30% 2706 32%
138 BugChess2 1.7 x64 2563 18 18 1280 26% 2741 29%
139 ChessTiger 2007 1.0012 w32 2562 12 12 2709 27% 2734 31%
140 Scorpio 2.6 JA x64 2556 18 18 1200 27% 2728 31%
141 Crafty 23.2 JA x64 2556 18 19 1120 28% 2720 30%
142 Chronos 1.99 x64, no ponder 2554 18 18 1160 28% 2716 33%
143 Crafty 23.3 JA x64, no ponder 2549 20 20 1000 25% 2741 30%
144 Tornado 4.25 x64 2541 19 20 1080 27% 2717 25%
145 Daydreamer 1.75 JA x64 2519 18 19 1200 25% 2709 30%
146 Tornado 3.67 x64 2479 20 21 1080 21% 2717 25%
147 Zarkov 6.44 w32 2473 15 15 2080 22% 2700 24%
148 Gaviota 0.80 x64 2358 25 26 1040 11% 2716 15% Frank
-
Frank Quisinsky
- Posts: 6824
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Hi there,
it was the most interesting compare I ever played.
On 4 of my 6 Q9550 systems SF 2.1.1 JA x64 default and SF 2.1.1 JA x64 PHQ runs to the same time. So we can compare an very interesting setting with default. On my two others Q9550 I do the same with ponder = off.
Result = PHQ is 10 ELO stronger after 1.200 games as default after 1.200 games.
The same test I make with ponder = off on the other two Q9550 systems I have.
Result = PHQ is 12 ELO stronger after 1.200 games as default after 1.200 games.
More interesting as the results are the games. The SWCR ponder = on games can be download on my webpage.
Look here ...
PHQ, 1.200 games SWCR ponder = on
Note: 8% ... clearly the NEW SWCR record !!!!
96 games won with mate up to move 55
9 games lost with mate up to move 55
PHQ, 1.200 games, ponder = off (my test systems)
91 games won with mate up to move 55
9 games lost with mate up to move 55
Default, 1.200 games, ponder = on
53 games won with mate up to move 55
3 games lost with mate up to move 55
Default, 1.200 games, ponder = off
49 games won with mate up to move 55
7 games lost with mate up to move 55
Ponder = on, PHQ / Default = 2.400 games, 14x lost on time
Ponder = off, PHQ / Default = 2.400 games, 00x lost on time
The PHQ (Pohl, Hartwig, Quisinsky) settings do a great job. I think we can make it a bit better with the following setting and the time controls I used in SWCR.
Could be PHQ2
Mobility (Middle Game)=115
Mobility (Endgame)=85
Aggressiveness=135
Cowardice=85
I test a bit with positions from the 4.800 SF matches I played on my 6 machines. This setting could be wounderful and perhaps a good mixed. Not tested in games yet but with different test positions.
SF PHQ plays the most interesting chess I ever saw. Problems with PHQ I saw in endgames with 16-10 pieces. The PHQ setting is a bit to aggressive and SF lost some points. Also in the early middlegame ... to fast SF PHQ try to gave knight, bishops for an attack and game ended remis. Here I saw 3 or 4 games. But all in all ... this setting is great ... but I believe only for longer time controls or for tactical middlegame analyzes ... for tactical middlegame analyzes I think the best what we can used.
Best
Frank
it was the most interesting compare I ever played.
On 4 of my 6 Q9550 systems SF 2.1.1 JA x64 default and SF 2.1.1 JA x64 PHQ runs to the same time. So we can compare an very interesting setting with default. On my two others Q9550 I do the same with ponder = off.
Result = PHQ is 10 ELO stronger after 1.200 games as default after 1.200 games.
The same test I make with ponder = off on the other two Q9550 systems I have.
Result = PHQ is 12 ELO stronger after 1.200 games as default after 1.200 games.
More interesting as the results are the games. The SWCR ponder = on games can be download on my webpage.
Look here ...
PHQ, 1.200 games SWCR ponder = on
Note: 8% ... clearly the NEW SWCR record !!!!
96 games won with mate up to move 55
9 games lost with mate up to move 55
PHQ, 1.200 games, ponder = off (my test systems)
91 games won with mate up to move 55
9 games lost with mate up to move 55
Default, 1.200 games, ponder = on
53 games won with mate up to move 55
3 games lost with mate up to move 55
Default, 1.200 games, ponder = off
49 games won with mate up to move 55
7 games lost with mate up to move 55
Ponder = on, PHQ / Default = 2.400 games, 14x lost on time
Ponder = off, PHQ / Default = 2.400 games, 00x lost on time
The PHQ (Pohl, Hartwig, Quisinsky) settings do a great job. I think we can make it a bit better with the following setting and the time controls I used in SWCR.
Could be PHQ2
Mobility (Middle Game)=115
Mobility (Endgame)=85
Aggressiveness=135
Cowardice=85
I test a bit with positions from the 4.800 SF matches I played on my 6 machines. This setting could be wounderful and perhaps a good mixed. Not tested in games yet but with different test positions.
SF PHQ plays the most interesting chess I ever saw. Problems with PHQ I saw in endgames with 16-10 pieces. The PHQ setting is a bit to aggressive and SF lost some points. Also in the early middlegame ... to fast SF PHQ try to gave knight, bishops for an attack and game ended remis. Here I saw 3 or 4 games. But all in all ... this setting is great ... but I believe only for longer time controls or for tactical middlegame analyzes ... for tactical middlegame analyzes I think the best what we can used.
Best
Frank
-
Frank Quisinsky
- Posts: 6824
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Note:
The mate statistic I wrote before is a bit ... I don't know the right words ... a bit with caution to see.
Different weaker engines have a bad king safty. Most of such games played SF vs. this group of engines. Also clear is that different programs are very strong in short win games. Hiarcs is a good example.
But both SF version must play vs. the same group of 30 engines the 40-game-matches. In this case the statistic is very interesting!
Best
Frank
And great is ...
Yes, to try out the SF parameters can be very interesting.
Congratualation to the SF team.
Default have 29 ELO as the predecessor 2.0.1 !!!
After a longer time ...
The mate statistic I wrote before is a bit ... I don't know the right words ... a bit with caution to see.
Different weaker engines have a bad king safty. Most of such games played SF vs. this group of engines. Also clear is that different programs are very strong in short win games. Hiarcs is a good example.
But both SF version must play vs. the same group of 30 engines the 40-game-matches. In this case the statistic is very interesting!
Best
Frank
And great is ...
Yes, to try out the SF parameters can be very interesting.
Congratualation to the SF team.
Default have 29 ELO as the predecessor 2.0.1 !!!
After a longer time ...
-
mcostalba
- Posts: 2684
- Joined: Sat Jun 14, 2008 9:17 pm
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Hi Frank,Frank Quisinsky wrote: Congratualation to the SF team.
Default have 29 ELO as the predecessor 2.0.1 !!!
After a longer time ...
congratulations to you for the PHQ setting, I will try it for sure, it is already on our todo list.
Yes, 29 ELO is more in line with our internal testing, and I see that also Ingo now shows a +22 ELO increase for 2.1.1, this is comfortable it means that our internal testing is not so totally broken as we guessed, anyhow for sure it can be improved.
Thanks again Frank for running the tournament !
Marco
-
Frank Quisinsky
- Posts: 6824
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Hi Marco,
after the latest release versions of SF I am a bit disappointed. SF 1.7.1 are more aggressive as all other newer versions. Think so ...
And now ...
The World isn't longer broken, can be so wonderful.
All is fine, SF is great and I can set without any problems the playing style I like. No longer blue clouds with sheeps ... Fire on the board is more interesting!
OK, default Stockfishes are great in aggressiveness too, but PHQ is clearly better, has other problems, default don't have
With other words.
All is OK with me
I like SF, Spark all the engines which play aggressive chess. Have more fun if I am looking in the still running games.
THANKS for this wounderful piece of software.
If you have 30 ELO more in your testing, I think all is OK. But honest, after 800 test games I had with 2.1 JA 6 ELO more too, could be random. I tested at first with the first compile release.
Again, many thanks ... If I can play violin ... you can be sure ... I would create a song for Tord, you and Joona.
All what we need is SF for my private GM middlegame analyzes. I mean I and my systems OK, Spark and Junior, Hiarcs and Hannibal and the othes too.
Best
Frank
PS: Don't need 1.7.1 any longer
2.1.1 is clear for take of on runway ... FIRE!
after the latest release versions of SF I am a bit disappointed. SF 1.7.1 are more aggressive as all other newer versions. Think so ...
And now ...
The World isn't longer broken, can be so wonderful.
All is fine, SF is great and I can set without any problems the playing style I like. No longer blue clouds with sheeps ... Fire on the board is more interesting!
OK, default Stockfishes are great in aggressiveness too, but PHQ is clearly better, has other problems, default don't have
With other words.
All is OK with me
I like SF, Spark all the engines which play aggressive chess. Have more fun if I am looking in the still running games.
THANKS for this wounderful piece of software.
If you have 30 ELO more in your testing, I think all is OK. But honest, after 800 test games I had with 2.1 JA 6 ELO more too, could be random. I tested at first with the first compile release.
Again, many thanks ... If I can play violin ... you can be sure ... I would create a song for Tord, you and Joona.
All what we need is SF for my private GM middlegame analyzes. I mean I and my systems
OK, Spark and Junior, Hiarcs and Hannibal and the othes too.Best
Frank
PS: Don't need 1.7.1 any longer
2.1.1 is clear for take of on runway ... FIRE!
-
Uri Blass
- Posts: 10427
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Here are the news from the IPON page(in the bottom of this post):mcostalba wrote:Hi Frank,Frank Quisinsky wrote: Congratualation to the SF team.
Default have 29 ELO as the predecessor 2.0.1 !!!
After a longer time ...
congratulations to you for the PHQ setting, I will try it for sure, it is already on our todo list.
Yes, 29 ELO is more in line with our internal testing, and I see that also Ingo now shows a +22 ELO increase for 2.1.1, this is comfortable it means that our internal testing is not so totally broken as we guessed, anyhow for sure it can be improved.
Thanks again Frank for running the tournament !
Marco
It seems that based on the IPON there may be 22 elo increase for 2.1.1 but only 12 elo increase for 2.1
I wonder if it is possible that the reason for the fact that you got different results is different speed.
Maybe you should have a function that calculates a number for stockfish's speed based on time that it needs to analyze some positions to fix depth
when you simply multiply the time by a constant in order to get always the same number in your hardware even with a different version.
If people with different hardware get a bigger number for new stockfish then it means that stockfish on their hardware is faster relative to your hardware and if they get smaller number for new stockfish then it means that stockfish on their hardware is slower.
Difference of 10% in speed can explain 10 elo difference so if people can got 5 for previous stockfish and 4.5 for new stockfish then they can expect 10 elo less on their hardware.
2011.06.14
Stockfish 2.1 replaced by Stockfish 2.1.1 (All 2400 games repeated). Minor
Elo increase (10 Elo).
2011.05.07
Games added - Stockfish 2.1 - 400 games
+1 Elo to the 2000 game result.
2011.05.06
Engine added - Stockfish 2.1 - 2000 games
An initial Elo increase of 7 points. Mainly a code cleaning and/or bug fixing.
-
mcostalba
- Posts: 2684
- Joined: Sat Jun 14, 2008 9:17 pm
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Yes it is. Speed on Intel hardware is the only difference between 2.1 and 2.1.1, on Intel hardware 2.1.1 is about 5-6% faster.Uri Blass wrote: I wonder if it is possible that the reason for the fact that you got different results is different speed.
Add to this some error bar noise and you can easily get the 10 ELO difference Ingo has experienced.
Uri Blass wrote:
Maybe you should have a function that calculates a number for stockfish's speed based on time that it needs to analyze some positions to fix depth
when you simply multiply the time by a constant in order to get always the same number in your hardware even with a different version.
Code: Select all
./stockfish bench-
IWB
- Posts: 1539
- Joined: Thu Mar 09, 2006 2:02 pm
Re: SWCR: SF 2.1.1 x64 default and PHQ, final and comments!
Hello
I doubt it is the speed difference. First of all 10% with some good luck we are talking about a maximum of 7 Elo. Then you mention 5-6% difference on Intel, which boils it down to a max of 4 Elo. These 4 Elo are well below everything I can measure. This will be completly lost in statistical noise.
Additionaly the IPON is running on AMD.
The bench result for the 2.1 is:
Total time (ms) : 6156
Nodes searched : 6487630
Nodes/second : 1053871
For the 2.1.1 it is:
Total time (ms) : 6016
Nodes searched : 6487630
Nodes/second : 1078395
the comparision might be difficult, but it look more like 2.3% then 6 ...
On my i7 (not used for the IPON) the difference is even lower, 1.8%.
My theory is much more the normal errorbar than speed differences or compiles.
Btw, the CEGT 40/20 has just 2 Elo from 2.0.1 to 2.1.1 on one core ...
Noise ... !
Bye
Ingo
I doubt it is the speed difference. First of all 10% with some good luck we are talking about a maximum of 7 Elo. Then you mention 5-6% difference on Intel, which boils it down to a max of 4 Elo. These 4 Elo are well below everything I can measure. This will be completly lost in statistical noise.
Additionaly the IPON is running on AMD.
The bench result for the 2.1 is:
Total time (ms) : 6156
Nodes searched : 6487630
Nodes/second : 1053871
For the 2.1.1 it is:
Total time (ms) : 6016
Nodes searched : 6487630
Nodes/second : 1078395
the comparision might be difficult, but it look more like 2.3% then 6 ...
On my i7 (not used for the IPON) the difference is even lower, 1.8%.
My theory is much more the normal errorbar than speed differences or compiles.
Btw, the CEGT 40/20 has just 2 Elo from 2.0.1 to 2.1.1 on one core ...
Noise ... !
Bye
Ingo