90-minute Engine Matches

[brinan]

Thought I would pass on results of my recent interest in watching computer chess matches. Discovered this forum only recently. Not saying I have watched all 114 games!

Time control has always been set to game in 90 minutes.

Bayeslo Data (covering 2nd Mar - 22nd Apr, 2010):

Code: Select all

Rank Name                Elo    +    - games score oppo. draws 
   1 IvanHoe 81          119  100   94    22   68%    21   64% 
   2 Igorrit 0086v7      116  255  232     4   63%    50   25% 
   3 Igorrit 0086v9       92  104  100    20   65%    14   70% 
   4 IvanHoe 63           68  203  216     4   38%   119   75% 
   5 FireBird 1.2         50   86   86    28   54%    30   64% 
   6 Stockfish 1.63       27   77   76    38   53%    10   53% 
   7 RobboLito 0.085g3     2   85   86    28   46%    17   71% 
   8 FireBird 1.1        -15  110  109    18   53%   -20   50% 
   9 IvanHoe 73          -40   95   98    24   40%    18   54% 
  10 Rybka 2.2n2         -50   85   89    32   38%    20   44% 
  11 Rybka 2.2mp         -66  275  326     2   25%     6   50% 
  12 Bright 0.5c        -114  230  251     4   38%   -37   25% 
  13 Cyclone xTreme     -189  232  294     4   25%   -55    0% 

GUI
Arena 2.0.1, 90 minutes each side, 256MB hash, ponder on, NEW Arena Mainbook 8.0.abk (max 12 half-moves), tablebases on

Engines
64 bit, 2 threads max

Hardware
Intel(R) Core(TM) i3-330M Processor (3M Cache, 2.13 GHz, 2 cores, 4 threads) with 2,936MB memory.
Fritz Chess Benchmarks:
3,849 KNPS [4 processors-to-use setting]
2,761 KNPS [2].

Operating System
Windows 7, Home Premium Edition (Build 7600)

PGN Download (MegaUpload site, ZIP file)

Notes:
1) Stockfish 1.63 used own book.
2) Not always had option to set number of threads or CPUs.
3) Estimated ratings,
with starting point http://computerchess.org.uk/ccrl/4040/r ... t_all.html ,
Stockfish 1.6.3 64-bit 2CPU 3112
Rybka 2.2n2 64-bit 2CPU 3056:

Code: Select all

IvanHoe 81	      3215
Igorrit 0086v7	  3212
Igorrit 0086v9	  3188
IvanHoe 63	      3164	
FireBird 1.2	    3146	
Stockfish 1.63	  3123  (3112 CCRL)
RobboLito 0.085g3  3098
FireBird 1.1	    3081
IvanHoe 73	      3056
Rybka 2.2n2	     3046  (3056 CCRL)
Rybka 2.2mp	     3030
Bright 0.5c	     2982	
Cyclone xTreme	  2907

[brinan]

Another 48 games.

Download all games. Previous link defunct.

Bayeslo Data (covering 2nd Mar - 11th May, 2010):

Code: Select all

Rank Name                Elo    +    - games score oppo. draws   Est ELO 
   1 IvanHoe 81          141   95   89    28   71%    21   57%      3243
   2 Igorrit 0086v9       84  100   96    22   64%    13   73%      3186
   3 IvanHoe 63           78  114  113    16   53%    62   69%      3180
   4 Igorrit 0086v7       72  121  118    16   56%    36   50%      3174
   5 IvanHoe T63Mini7     61  134  136    10   45%    80   90%      3163
   6 FireBird 1.2         33   76   76    38   51%    25   66%      3135
   7 Stockfish 1.7.1      26  112  115    20   45%    56   30%      3128
   8 Stockfish 1.63       22   69   69    48   51%    14   56%      3124
   9 RobboLito 0.09        5  124  127    14   43%    40   57%      3107
  10 RobboLito 0.085g3    -4   87   87    28   46%    11   71%      3098
  11 FireBird 1.1        -24  112  111    18   53%   -28   50%      3078
  12 IvanHoe 73          -44   97  100    24   40%    15   54%      3058
  13 Rybka 2.2n2         -57   87   90    32   38%    14   44%      3045
  14 Rybka 2.2mp         -74  279  331     2   25%    -1   50%      3028
  15 Bright 0.5c        -121  234  255     4   38%   -44   25%      2981
  16 Cyclone xTreme     -195  236  298     4   25%   -62    0%      2907

[brinan]

74 games added, making 236 in total. Table and games available via chess.brinan.me.uk, then clicking Ratings at the top.

From now on, I will only post about new tournaments, each in their own thread.

A new base rating has been used: 3012 for Rybka 2.2n2. As a result, I have adjusted the games file for the new ratings, and I have re-titled past tournaments to include category levels.

The new base rating came about when I started seeing how far I could link engine ratings to human ratings, no matter how questionable.

Stages I went through:
1. October 2006, Vladimir Kramnik rated 2750;
2. Kramnik lost 2-4 to Deep Fritz 10 (2 core, assume 32-bit), representing 125 Elo points superiority for Deep Fritz;
3. CEGT 40/120 ratings, Deep Fritz 254 points behind Rybka 3 x64 2CPU, giving Rybka 3 a rating of 3129;
4. CCRL 40/40 ratings, Rybka 2.2n2 125 points behind Rybka 3 x64 2CPU;
5. CCRL 40/40 equivalent to 40/35 on my laptop, so, say, 8 points extra for improved processing power (5 / 20 * Elo 30 [say]).

Links to various sites on my chess web pages.

All this leads to an expectation that top engines, running on my laptop, would beat Magnus Carlsen (2813) something like 21.5 to 2.5 in a 24-match contest!

Yes, I am being very fast and loose with data. However, I wanted the link. If others put forward their own weird and wonderful sequences, the ratings can always be recast.

[brinan]

New 6-game match started: Fire 1.31 v Igorrit 0086v9.

First game played. Enjoyed this one (writing as an untrained observer).

Did say I would start a new thread for each additional event. However, seems a waste of a good thread!

[Taner Altinsoy]

Why don't you use Stockfish 1.7.1 and Fire 1.31 and possibly other strong versions of Ivanhoe?

[brinan]

I do run them.

Bear in mind:
1. data collection is not my prime motive;
2. all games take a maximum of 3 hours, so it does take time to put results together for newer versions.

If you come across a stronger IvanHoe than those currently in my list, do let me know. You can find them by clicking Ratings at http://chess.brinan.me.uk/.

Regards.

[Taner Altinsoy]

Ups my bad didn't see stockfish 1.7.1.

I really appreciate the work you do. I wish I had a 64 bit system dedicated to computer chess. I only have a celeron 1400 32 bit where I run intensive tournaments.

Other engines you might consider are IvanhoeT63 mini10, Ivanhoemod12c, IvanhoeT63B, Houdini 1.01 and Firebird 1.1 with Dr.D settings (it is certainly much beter than default). But I completely understand the time it takes to add new engines to your list and would like to thank you again for your effort.

Taner

[brinan]

Fire 1.31 wins 4-2.

In game 6, both thought Igorrit up nearly 2 pawns deep into the middle game. But Fire won. Visit chess.brinan.me.uk to play through the game.

Also updated ratings data. All 248 games to date can be downloaded.

Ratings not generating the personal interest I thought they would. Sample size will never be sufficient, and pairings ought to be more distributed.

So, added a Ladder page. Still being thought out, so any suggestions welcome. A Ladder will allow flexibility and added, personal interest in the matches I arrange.

[brinan]

Two ladder matches started:
Houdini 1.02 v IvanHoe 63,
Fire 1.31 v IvanHoe 57.

One game in each already played. Matches fought over 12 games.

Have set up a RSS feed on the main chess webpage.

Access to all game data available by clicking on www button below.

[brinan]

Ladder match challenges over.

IvanHoe 63 loses its rank to Houdini 1.02 (4.5 - 6.5). Houdini now ranked 1.
IvanHoe 57 defends challenge by Fire 1.31 (6 - 4). IvanHoe 57 remains at rank 3.