Winboard Forum

http://greko.chess.googlepages.com/greko-5.2-win.zip
http://greko.chess.googlepages.com/greko-5.2-linux.zip

Version 5.2
12-Feb-2007
-----------

(fix) Main hash table size is not divided by 2 anymore.

(new) Added futility pruning, in a very simple and straightforward form.

(chg) Search: tuned some conditions for LMR and Nullmove.
(chg) Eval: King safety value scaled up a little.

Overall improvement in strength is 50...70 points for blitz games.

Vladimir Medvedev wrote:http://greko.chess.googlepages.com/greko-5.2-win.zip
http://greko.chess.googlepages.com/greko-5.2-linux.zip

Version 5.2
12-Feb-2007
-----------

(fix) Main hash table size is not divided by 2 anymore.

(new) Added futility ptuning, in a very simple and straightforward form.

(chg) Search: tuned some conditions for LMR and Nullmove.
(chg) Eval: King safety value scaled up a little.

Overall improvement in strength is 50...70 points for blitz games.

Your engine seem to be a threat to Rybka at this rate.

Tony Thomas wrote:Your engine seem to be a threat to Rybka at this rate.

Simple plot in Excel shows me that GreKo is somewhere near it's asymptotic saturation, at the level of maybe 2400...

Vladimir Medvedev wrote:

Tony Thomas wrote:Your engine seem to be a threat to Rybka at this rate.

Simple plot in Excel shows me that GreKo is somewhere near it's asymptotic saturation, at the level of maybe 2400...

It was already around 2350 with the previous version, now with this improvement you probably broke the 2400 barrier. You are the programmer, you know about your program better than me.

Maybe 50...70 points was overestimation. In my toy rating pool the difference is smaller.
http://greko.chess.googlepages.com/openrating

Code: Select all

Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Toga II 1.2.1a                 : 2731   89  84    84    86.3 %   2411   17.9 %
  2 Fruit 2.1                      : 2700   98  92    84    83.9 %   2413    8.3 %
  3 Scorpio 1.8                    : 2600   76  73    84    73.8 %   2420   19.0 %
  4 SlowChessBlitz 2.96            : 2558   75  72    84    68.5 %   2423   15.5 %
  5 Delfi 5.0                      : 2554   72  70    84    67.9 %   2424   19.0 %
  6 Glaurung 1.2.1                 : 2549   73  71    84    67.3 %   2424   17.9 %
  7 Crafty 20.14                   : 2545   77  74    84    66.7 %   2424    9.5 %
  8 Phalanx XXII                   : 2425   69  69    84    48.8 %   2433   16.7 %
  9 GreKo 5.2                      : 2354   72  73    78    37.2 %   2445   17.9 %
 10 EXchess 5.0.1 beta             : 2350   69  71    84    37.5 %   2438   17.9 %
 11 GreKo 5.1                      : 2330   76  78    78    34.0 %   2445   11.5 %
 12 GreKo 2.817                    : 2320   74  77    84    33.3 %   2440    9.5 %
 13 GNUChess 4 TM                  : 2307   76  79    84    31.5 %   2441    8.3 %
 14 HoiChess 0.6.1                 : 1985  149 121    84     6.0 %   2464    0.0 %
 15 TSCP 1.81                      : 1967  153 101    84     5.4 %   2466    1.2 %

I had chosen Greko as a sparring partner for Romichess, but I changed my mind later because Greko was getting stronger at an alarming rate, and I was looking for two engines weaker than Romi. It looks to me after around 180 games that the current version of Romi is way too strong for the weaker engines that I picked, so I am going to let Greko feel the heat from Romi.

New update:

http://greko.chess.googlepages.com/greko-5.2.5-win.zip
http://greko.chess.googlepages.com/grek ... source.zip

You release versions so fast that I barely have time to play with them. Here is my private database that I created to test future versions of Romichess. Its not finished yet, but its almost done. Note that Zappa 1.1's rating is fixed at 2600. Provided that it is at least a 2700 engine, add about 100 points, you get the realistic rating of some of the engines.

2/23/2007 3:19:50 PM :

Program Elo + - Games Score Av.Op. Draws

Code: Select all

  1 Zappa 1.1                      : 2600   36  35   360    71.5 %   2440   16.4 %
  2 Arasan 9.5                     : 2558   35  34   360    66.0 %   2443   15.3 %
  3 Francesca MAD 0.13             : 2550   34  33   360    64.9 %   2444   19.2 %
  4 Danasah 2.85                   : 2498   32  32   360    57.1 %   2448   20.8 %
  5 Phalanx Reborn                 : 2475   33  33   360    53.6 %   2450   16.7 %
  6 RomiChessDK2                   : 2475   33  33   360    53.6 %   2450   16.1 %
  7 Delphil 1.6c                   : 2455   37  37   284    48.2 %   2467   18.3 %
  8 NanoSzachy 2.7                 : 2433   41  41   233    40.1 %   2502   19.3 %
  9 Djinn 0.925x                   : 2417   39  39   270    42.0 %   2473   15.2 %
 10 Horizon 4.3                    : 2415   44  44   210    36.4 %   2512   17.6 %
 11 Zeus 1.28                      : 2409   33  33   360    43.3 %   2456   17.2 %
 12 GreKo 5.2                      : 2347   45  46   219    28.3 %   2509   16.4 %
 13 Lime 6.3                       : 2214   50  51   270    16.5 %   2495   10.7 %

Tony Thomas wrote:You release versions so fast that I barely have time to play with them. Here is my private database that I created to test future versions of Romichess. Its not finished yet, but its almost done. Note that Zappa 1.1's rating is fixed at 2600. Provided that it is at least a 2700 engine, add about 100 points, you get the realistic rating of some of the engines.

2/23/2007 3:19:50 PM :

Program Elo + - Games Score Av.Op. Draws

Code: Select all

  1 Zappa 1.1                      : 2600   36  35   360    71.5 %   2440   16.4 %
  2 Arasan 9.5                     : 2558   35  34   360    66.0 %   2443   15.3 %
  3 Francesca MAD 0.13             : 2550   34  33   360    64.9 %   2444   19.2 %
  4 Danasah 2.85                   : 2498   32  32   360    57.1 %   2448   20.8 %
  5 Phalanx Reborn                 : 2475   33  33   360    53.6 %   2450   16.7 %
  6 RomiChessDK2                   : 2475   33  33   360    53.6 %   2450   16.1 %
  7 Delphil 1.6c                   : 2455   37  37   284    48.2 %   2467   18.3 %
  8 NanoSzachy 2.7                 : 2433   41  41   233    40.1 %   2502   19.3 %
  9 Djinn 0.925x                   : 2417   39  39   270    42.0 %   2473   15.2 %
 10 Horizon 4.3                    : 2415   44  44   210    36.4 %   2512   17.6 %
 11 Zeus 1.28                      : 2409   33  33   360    43.3 %   2456   17.2 %
 12 GreKo 5.2                      : 2347   45  46   219    28.3 %   2509   16.4 %
 13 Lime 6.3                       : 2214   50  51   270    16.5 %   2495   10.7 %

I know no list in which Zappa on a single machine is anywhere
near 2700 in relation to what you list here.

Guenther

Guenther Simon wrote:I know no list in which Zappa on a single machine is anywhere
near 2700 in relation to what you list here.

Guenther

Code: Select all

CCRL 40/40 Rating List - 32-bit single-CPU engines
48877 games played by 254 programs, run by 13 testers
Ponder off, General books (up to 12 moves), 3-4-5 piece EGTB
Time control: Equivalent to 40 moves in 40 minutes on Athlon 64 X2 4600+ (2.4 GHz)
Computed on February 24, 2007 with Bayeselo based on 48'877 games
Tested by CCRL team, 2005-2007, http://computerchess.org.uk/ccrl/4040/

 Rank                Engine                 ELO   +    -   Score  AvOp  Games
    1 Rybka 2.2 32-bit                     2995  +21  -21  72.2% -153.1   819
      Rybka 1.2 32-bit                     2977  +18  -18  70.7% -149.3  1075
      Rybka 2.1 32-bit                     2970  +23  -23  69.5% -130.6   646
      Rybka 2.1 32-bit (C+1 OSO)           2966  +31  -30  72.3% -160.5   410
      Rybka 1.1 32-bit                     2960  +18  -17  69.7% -138.5  1162
      Rybka 2.3 32-bit                     2940  +85  -83  59.0%  -51.0    39
      Zap!Chess Zanzibar 32-bit            2889  +51  -51  54.1%  -26.4   109
      Rybka 1.0 32-bit                     2884  +16  -16  60.7%  -72.7  1240
    2 Hiarcs 11                            2876  +29  -28  54.9%  -29.4   379
    3 Fritz 10                             2873  +33  -33  55.6%  -38.4   296
    4 Shredder 10                          2872  +16  -15  60.7%  -72.1  1392
    5 Loop 10.32f                          2871  +19  -19  51.0%   -5.8   841
      Hiarcs X50 Hypermodern               2864  +28  -28  50.2%   -1.7   419
    6 Toga II 1.2.1a                       2862  +13  -13  49.6%   +0.6  2010
      Loop 12.32                           2856  +37  -36  56.6%  -44.5   234
  7-8 Naum 2.1 32-bit                      2850  +32  -32  54.4%  -32.3   306
  7-8 Spike 1.2 Turin                      2850  +13  -13  51.1%   -8.5  2002
      Toga II 1.2 32-bit                   2846  +32  -32  55.5%  -38.1   309
      Hiarcs X50                           2845  +17  -18  45.8%  +30.5  1114
      Fritz 9                              2842  +13  -13  50.0%   +0.8  2111
      Hiarcs X54 Hypermodern 32-bit        2841  +40  -40  49.3%   +5.0   203
      Fruit 2.2.1                          2840  +13  -13  52.8%  -17.7  1973
      Hiarcs 10                            2839  +15  -15  50.1%   +0.8  1494
    9 Junior 10                            2839  +18  -18  50.3%   -2.5  1006
      Hiarcs X54 32-bit                    2828  +36  -36  45.3%  +34.2   247
      Toga II 1.1a                         2826  +18  -18  55.2%  -34.3  1017
      Chess Tiger 2007 Gambit              2823 +349 -239  90.0% -240.8     5
      Shredder 9                           2823  +17  -17  55.7%  -38.3  1162
      Junior 10.1                          2821  +93  -93  51.3%   -9.5    39
      Hiarcs 10 Hypermodern                2818  +19  -19  51.7%   -8.4   889
      Hiarcs 11.1                          2816 +170 -170  50.0%  +13.9     9
      Scorpio 1.82                         2814 +183 -191  43.8%  +36.0     8
   10 Ktulu 8.0                            2803  +17  -17  43.2%  +45.3  1163
      Scorpio 1.91                         2801  +61  -60  60.0%  -63.2    90
      Fritz 8 Bilbao                       2800  +27  -27  44.9%  +37.1   442
      Naum 2.0 32-bit                      2800  +18  -18  51.0%   -4.1   959
   11 Chess Tiger 2007                     2795  +18  -18  46.5%  +23.3   961
      Spike 1.1                            2786  +15  -15  49.0%   +6.2  1469
      Scorpio 1.84                         2784  +53  -53  56.7%  -71.4   141
      Scorpio 1.9                          2783  +55  -55  50.0%   -6.7   114
   12 Zap!Chess Paderborn 32-bit           2781  +21  -21  45.2%  +34.5   732
      Junior 9                             2780  +16  -16  46.6%  +21.7  1237
      Glaurung 1.2.1 Avalanche 32-bit      2772  +48  -48  52.4%  -17.7   147
      Zap!Chess Paderborn SE 32-bit        2772  +36  -36  45.7%  +28.4   245
      Fruit 2.1                            2770  +49  -48  54.7%  -42.0   139
   13 Glaurung 1.2.1 Crusader 32-bit       2765  +41  -41  52.2%  -20.2   208
      Zap!Chess Reykjavik 32-bit           2765  +35  -35  37.8%  +78.5   255
   14 CM10th Paralyse                      2764  +25  -25  42.6%  +49.3   535
      Ktulu 7.5                            2762  +16  -16  42.9%  +48.5  1421
      Glaurung 1.2.1 32-bit                2761  +17  -17  45.5%  +31.5  1144
      CM9000 Enforcer                      2760  +33  -33  49.0%   +4.4   291
      CM10th Xperience                     2759  +16  -16  42.0%  +54.5  1404
      LoopList 6.00                        2759  +43  -43  54.3%  -43.6   184
   15 Smarthink 1.00 32-bit                2756  +16  -16  43.5%  +42.9  1304
      Spike 1.0a                           2752  +19  -19  43.9%  +41.2   871
      CM10th Magic II                      2746  +17  -18  40.6%  +61.4  1037
   16 Scorpio 1.8                          2746  +19  -19  47.9%  +14.7   877
      CM10th Lazarus                       2743  +41  -41  49.8%   +1.4   200
      CM10th Ogre                          2743  +35  -35  47.1%  +18.0   258
   17 Slow Chess Blitz WV2.1               2741  +16  -16  48.8%   +8.5  1250
      CM10th Warrior                       2738  +38  -38  46.3%  +28.0   219
      Ktulu 7.1                            2738  +19  -19  40.9%  +62.9   949
      Chess Tiger 15 Gambit                2735  +31  -31  45.2%  +37.1   347
      CM10th Default                       2733  +17  -18  40.0%  +68.0  1089
      Chess Tiger 15                       2733  +17  -17  41.3%  +59.9  1129
      CM10th Pestilence                    2729  +35  -35  49.4%   +2.5   266
      Glaurung 1.1 32-bit                  2729  +38  -39  45.9%  +31.4   231
   18 WildCat 7                            2727  +19  -19  45.1%  +33.9   895
      Glaurung 1.0.2 32-bit                2724  +18  -18  42.3%  +56.0  1072
   19 Gandalf 6                            2723  +16  -16  40.0%  +68.7  1411
20-21 Pro Deo 1.2                          2722  +22  -22  49.7%   -1.5   712
20-21 Ruffian 2.1.0                        2722  +26  -26  43.5%  +50.1   509
      List 5.12                            2720  +20  -20  46.4%  +23.3   858
      Hiarcs 9                             2717  +20  -21  37.8%  +82.1   784
      Chess Tiger 2004                     2716  +22  -22  40.4%  +61.9   653
      Delfi 5.1                            2713 +236 -212  70.0% -108.8     5
      Ruffian 1.0.5                        2713  +21  -21  45.4%  +29.7   732
      Scorpio 1.7                          2711  +36  -36  54.5%  -32.1   245
      Slow Chess Blitz WV2                 2711  +19  -19  40.1%  +69.0   950
   22 Pharaon 3.5.1                        2707  +18  -18  43.6%  +44.9   998
   23 Delfi 5.0                            2703  +20  -20  47.5%  +15.5   775
      Naum 1.91 32-bit                     2703  +18  -18  39.3%  +72.5  1042
   24 Aristarch 4.50                       2699  +16  -16  42.7%  +52.0  1335
      WildCat 6                            2697  +22  -22  47.1%  +18.6   680
      Delfi 4.6                            2696  +31  -31  50.0%   -2.3   334
   25 Pseudo 0.7c                          2693  +21  -21  49.5%   +2.7   707
      Pro Deo 1.1                          2690  +32  -32  45.7%  +30.9   323
      Crafty 21.5 32-bit                   2689 +105 -100  63.6%  -90.7    33
      Zappa 1.1 32-bit                     2686  +40  -40  47.6%  +18.3   207

Graham Banks wrote:

Guenther Simon wrote:I know no list in which Zappa on a single machine is anywhere
near 2700 in relation to what you list here.

Guenther

The key word was 'in relation'. Everyone can create a list,
where Zappa 1.1 has 7000 rating points whatever it means.
The ratings at CCRL are somehow inflated against all other
rating lists by 60-90 points. Obviously you scale different than
most others, may be also a consequence of yet much too less
entries below '2600' in your list?
In the real world Kiwi is not considered having 2567 rating
points anyway comparable to something like Elo or consensus
by other testers for the last years.
You cut off the part which could be compared with Tonys'...
(BTW you should consider testing FrancescaMad, it seems to miss
there?)

Code: Select all

Zappa 1.1 32-bit   2686   +40   −40   47.6%   +18.3   32.4%   207
Arasan 9.5         2560   +43   −44   47.1%   +19.6   24.3%   189
DanaSah 2.85       2536   +52   −52   49.6%   +7.6    29.8%   124


Code: Select all

  1 Zappa 1.1                      : 2600   36  35   360    71.5 %   2440   16.4 %
  2 Arasan 9.5                     : 2558   35  34   360    66.0 %   2443   15.3 %
  3 Francesca MAD 0.13             : 2550   34  33   360    64.9 %   2444   19.2 %
  4 Danasah 2.85                   : 2498   32  32   360    57.1 %   2448   20.8 %

Regards,
Guenther

Guenther Simon wrote:(BTW you should consider testing FrancescaMad, it seems to miss
there?)
Guenther

Hi Guenther,

we can't test Francesca, Bruja or Tytan because their internal opening books can't be disabled.
We don't allow engines to use their own books.

We can't test RomiChess either because learning can't be disabled (although Michael has said he'll make turning learning off an option at some stage).

Regards, Graham.

According to Leo's rating list Zappa(take off 40 points since he is using 64bit) is a 2700 engine, when I said add 100 points, I said some engines and I was refering mainly to engines towards the end of the list. There rating is rather skewed. He didnt test the same versions of other engines so...

Code: Select all

17 Zappa 1.1 x64             2714   42   42   184   51%  2711   34%
47 FranMAD 0.13              2567   51   51   124   60%  2502   35%
70 Djinn 0.925x              2476   62   60    92   66%  2364   30%
76 Phalanx XXII              2455   18   18  1084   48%  2467   25%

Graham Banks wrote:

Guenther Simon wrote:(BTW you should consider testing FrancescaMad, it seems to miss
there?)
Guenther

Hi Guenther,

we can't test Francesca, Bruja or Tytan because their internal opening books can't be disabled.
We don't allow engines to use their own books.

We can't test RomiChess either because learning can't be disabled (although Michael has said he'll make turning learning off an option at some stage).

Regards, Graham.

It seems that you are not interested in work arounds. You can just delete the learnfile of Romichess after each 2 games. I know that is something hard to do, if I had more memory I would have helped you out with getting around two or three hundred games of Romi. Also, why not test Bruja, Tytan and Francesca and add on the name tag to specify that they are using own book? Also, when using a GUI book, the role of the own book is very small because none of those programs got a big book.

Tony Thomas wrote:It seems that you are not interested in work arounds. You can just delete the learnfile of Romichess after each 2 games. I know that is something hard to do, if I had more memory I would have helped you out with getting around two or three hundred games of Romi. Also, why not test Bruja, Tytan and Francesca and add on the name tag to specify that they are using own book? Also, when using a GUI book, the role of the own book is very small because none of those programs got a big book.

Sorry - no compromises.
We test with learning off and with no engine using its own book - no exceptions.

Version 5.3
20-May-2007
-----------

(chg) Using hash eval from previous iteration for pruning
(chg) Small optimization in attack detector
(chg) Small optimization in PV handling
(fix) Fixed crash on epd-files ended with empty lines

http://greko.chess.googlepages.com/greko-5.3-win.zip
http://greko.chess.googlepages.com/greko-5.3-linux.zip
http://greko.chess.googlepages.com/greko-5.3-source.zip

Enjoy!

Vladimir Medvedev wrote:Version 5.3
...
greko-5.3-linux.zip
...
greko-5.3-source.zip

Thank you Vladimir!

I just compared the time neede for "perft 5" in UVI mode here on my Core2Duo E6300:

Linux executable delivered: 1.34 seconds
Own compilation using the Makefile delivered: 0.92 seconds
Compiled with -O3 option: 0.78 seconds.

Maybe others can make some tests too.

Volker

Thank you, Volker!

I use gcc 3.3.3 under SuSE Linux. But I never test or profile GreKo under Linux, my development platform is Windows.
Can I use your Linux compilation as "official" one and put it on GreKo's page?

Vladimir Medvedev wrote:Thank you, Volker!

I use gcc 3.3.3 under SuSE Linux. But I never test or profile GreKo under Linux, my development platform is Windows.
Can I use your Linux compilation as "official" one and put it on GreKo's page?

Sure here it is:http://www.vpittlik.org/temp/greko-5.3.bz2 . No guarantee of any kind of course! I'm also running SuSE but gcc version is 4.1.2.

Thank you Volker for your Linux version. On my Linux box it works +24% faster than original (on perft 6 task). And version compiled by gcc 3.3.3 with -O3 surprisingly is -15% slower than original.