FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Guenther
Posts: 3676
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Guenther » Tue Jan 12, 2021 10:18 pm

xr_a_y wrote:
Tue Jan 12, 2021 9:51 pm
Thanks for reporting.

Promoting to n+b is often seen from minic because KBNvK is declared winning while KQPK or KQBK is just a big material advantage if mate is not found immediately. No worries about that.

But not winning the KBNvK is a real issue of course. Today at CCC minic even draw a KRvK. I am working on the issue currently.
Please look also at move

Code: Select all

63. g7 -M284/121 0s
from the original pgn.
Here either Minic suddenly changed the sign of the eval or the Shredder GUI had a hickup?

(Actually the Shredder GUI has some problems with converting near mate scores to something readable,
that was the main reason, why I had to modify/correct the Shredder PGN from Frank, to make the new tools work.
Sometimes though it's probably the engines themselves, which give strange output, but I have to debug
this first in other GUIs. I have seen things like +M0 +M-xxxx -M-xxxx and others in the pgn for a few programs)

Here are the first 820 games (first full round - same order as pgn) graphs as a shared photo book!
https://photos.google.com/share/AF1QipM ... A3ZVJWTW13

Frank Quisinsky
Posts: 5411
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Full name: Frank Quisinsky
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Frank Quisinsky » Tue Jan 12, 2021 10:24 pm

Hi Guenther,

nice graphics you create here for each of the games (round 1 out of 50).

Thank you, I will looking later ...
At the moment I a bit work if I have to replaced directly 4 engines.

Best
Frank
I like computer chess!

Guenther
Posts: 3676
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Guenther » Tue Jan 12, 2021 10:28 pm

Frank Quisinsky wrote:
Tue Jan 12, 2021 10:24 pm
Hi Guenther,

nice graphics you create here for each of the games (round 1 out of 50).

Thank you, I will looking later ...
At the moment I a bit work if I have to replaced directly 4 engines.

Best
Frank
Just a last little hint in case someone does not know. If you click on a picture it will be enlarged and there are arrows to go through
the whole collection, as usual with a 'gallery'.

User avatar
xr_a_y
Posts: 1401
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by xr_a_y » Wed Jan 13, 2021 6:44 am

So in addition to the strange end game behavior, you also have Minic loosing by time?
It is probably more a "connection stalled" because Minic crashed rather than a real loos by time forfeit.
I guess these happens very near mate or even at mate on loosing positions?

In the case (but I don t really think it is) it is really a time issue, there is a UCI parameter to increase a security margin in recent Minic versions (default is 50ms).

I am tracking down the issue currently is it appears also is CCC testing. I suspect this may happen or happen more maybe when using more thread than physical cores. I don t think this is good practice for chess engine testing. But anyway, even if it s the case, this is still a bug on my side.

Thanks for reporting.

Frank Quisinsky
Posts: 5411
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Full name: Frank Quisinsky
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Frank Quisinsky » Wed Jan 13, 2021 8:21 am

Hi,

it is not directly a crash by your engine or the Shredder GUI.
Minic lost on time, do nothing 2-6 moves before mate.
And Shredder GUI will not start the next game with the result ...
9 of 10 matches are still running, I have to restart the game.
The problem here, I am not sitting 24 hours on the system and lost a lot of time.

For me important that all engines runs without such problems or the tournament is to many work.

Other strong engines, like ChessBrainVB, Beef and some others have the same problem.
I don't know why all the years so many chess engines produced again and again the same problem.
Since more as 20 years an error in many chess programs.

Best
Frank
I like computer chess!

Kieren Pearson
Posts: 59
Joined: Tue Dec 31, 2019 1:52 am
Full name: Kieren Pearson

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Kieren Pearson » Wed Jan 13, 2021 10:29 pm

Guenther wrote:
Tue Jan 12, 2021 8:16 pm
Thanks for the tournament! It came right in time for testing some new to come tools.

Here Minic stranglely could not win from this position and also converted both pawns in the most difficult winning endgame.
Lucky Halogen didn't waste any time anymore since being lost and played all moves since move 54 in 0 seconds.
And later it allowed a 50 moves draw while permanently showing mate scores with K,B.N vs. K.



modified pgn


(right click on the picture and 'show image' will show full size)
Image

original pgn
The tournament used syzygy tablebases I'm guessing right? If Halogen is in a position that's in the tablebase at the root of the search it will just play the TB move instantly. Did Minic not have access to TB or does Minic not implement them?

User avatar
xr_a_y
Posts: 1401
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by xr_a_y » Thu Jan 14, 2021 6:32 am

Minic can use TB. Implementation of them are Arasan styled.

Frank Quisinsky
Posts: 5411
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Full name: Frank Quisinsky
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Frank Quisinsky » Fri Jan 15, 2021 4:46 am

Hi Vivien,

yes, in most cases it have to do with code from endgame bases.
I am sorry for disqualified your engine but during the toruney is still running I lose to many time.

I hope you can fix the problem for the next tourney.

Best
Frank

I wonder because Arasan never have the "Lost on time" problems.
A good example for "Lost on time" problems is Gullchess 3.1 (same problems with the syzygy code, many programmers used).
I like computer chess!

Frank Quisinsky
Posts: 5411
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Full name: Frank Quisinsky
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Frank Quisinsky » Fri Jan 15, 2021 4:51 am

Hi there,

now the torunament files are corrected.
I made in detail 4 changes, my News-27:

January 13th, 2021 (15:30) News-27
*** Minic 3.02 NaNi Skylake x64 = disqualified ***


After three x lost on time in round 3 out of 50 I disqualified Minic 3.02 NaNi Skylake x64.

I had the same problems with Minic 2.33 x64 for FCP Toruney-2020 and reported the bug. Minic lost often
on time around 2-6 moves before own mate (Shredder 13 GUI information: Analyze = empty).

That is s bad situation because Klaus and I must make the statistics for round 1-2 again. I did not test
Minic before I started the tournament. The problem is homemade!

OK, since I started the tournament two interesting engine updates are available. I added the updates. In
additional the current Stockfish dev. version.

01. Minic 3.02 NaNi Skylake will be replaced with Hannibal 1.7 x64.
02. SlowChess Blitz 2.4 NN x64 is replaced with SlowChess 2.5 NN AVX2 x64.
03. Marvin 4.0.1 POPCNT x64 is replaced with Marvin 5.0.0 NN AVX2 x64.
04. Sf 2020-12-31_18 NN BMI2 x64 is replaced with Stockfish 110121 NN BMI2 x64.
Timestamp: 1610390551, Stockfish is running in Hybrid mode (default).

With Stockfish I am not sure?
I suppose NNUE will be load with Shredder 13 GUI only, if the full path for the NNUE file in UCI
configuration is given. Furthermore, the nn-62ef826d1a6d.nnue file should be add in Stockfish directory.

My configurations can be found under:
http://www.amateurschach.de/main/_configuration.htm

462 games are to replay, these need ~ 2 days!

---

Now all games replayed, all statistics files are corrected and updated!
Results from Round 03 out of 50 are online.

We have a fight for place 1 (Stockfish vs. Komodo).
We have a fight for place 4 (Ethereal vs. SlowChess).
We have a fight for place 20 (Wasp vs. Gullchess, Arasan and Winter I believe, most interesting for myself).

A good start for Marvin 5.0.0 NN AVX2 x64 (seems that Wasp have problems vs. Marvin, will be see with the next games).

Best
Frank
I like computer chess!

Frank Quisinsky
Posts: 5411
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Full name: Frank Quisinsky
Contact:

Re: FCP Tourney-2021, 41.000x 45min. games on i9-10900k, 13xNN, now still running!

Post by Frank Quisinsky » Fri Jan 15, 2021 4:53 am

After the corrections!

Code: Select all

  FCP Tourney-2021
  ----------------

  Date           : January 15th, 2021 (05:00)
  Games          : 02.460
  Version        : 03.0, after R03
  Elo            : Shredder 13 POPCNT x64 = 3125
  Conditions     : http://www.amateurschach.de/main/_fcp-tourney-2021.htm

  Generated with : Ordo 1.2.6 x64 (Ratings for chess and other games)
  https://sites.google.com/site/gaviotachessengine/ordo
  by Miguel A. Ballicora (Argentina)



   # Player                           :      Elo  Games  Score%   won  draw  lost  Points  Draw%   Error   OppAvg   OppE  MoveAvg
  -------------------------------------------------------------------------------------------------------------------------------
  01. Dragon by Komodo NN AVX2 x64    :  3510.37    120    89.6    95    25     0   107.5   20.8   73.46  3096.96  42.99     80.8
  02. Stockfish 110121 NN BMI2 x64    :  3487.21    120    88.3    92    28     0   106.0   23.3   69.83  3097.54  43.08     80.0
  03. Ethereal 12.75 PEXT x64         :  3298.14    120    73.3    61    54     5    88.0   45.0   45.29  3102.27  43.69     97.4
  04. SlowChess 2.5 NN AVX2 x64       :  3290.41    120    72.5    62    50     8    87.0   41.7   47.42  3102.46  43.64     85.7
  05. Nemorino 6.04 NN PEXT x64       :  3225.79    120    65.4    50    57    13    78.5   47.5   42.73  3104.08  43.76     86.8
  06. RubiChess 1.9 NN BMI2 x64       :  3222.62    120    65.0    46    64    10    78.0   53.3   44.15  3104.16  43.72     83.5
  07. Xiphos 0.6 BMI2 x64             :  3205.10    120    62.9    43    65    12    75.5   54.2   43.44  3104.60  43.74     93.8
  08. Igel 2.9.0 NN BMI2 x64          :  3187.96    120    60.8    41    64    15    73.0   53.3   42.52  3105.02  43.76     86.7
  08. rofChade 2.3 BMI2 x64           :  3187.66    120    60.8    45    56    19    73.0   46.7   42.93  3105.03  43.75     89.7
  10. Defenchess 2.3 dev BMI2 x64     :  3174.32    120    59.2    45    52    23    71.0   43.3   41.88  3105.37  43.78     93.1
  -------------------------------------------------------------------------------------------------------------------------------
  11. Pedone 3.0 NN BMI2 x64          :  3171.30    120    58.8    36    69    15    70.5   57.5   42.37  3105.44  43.77     88.2
  12. Lc0 0.26.3 NN CPU x64           :  3167.75    120    58.3    38    64    18    70.0   53.3   40.41  3105.53  43.81     93.9
  12. Laser 1.7 BMI2 x64              :  3167.71    120    58.3    37    66    17    70.0   55.0   40.90  3105.53  43.80     88.6
  14. Booot 6.4 POPCNT x64            :  3161.16    120    57.5    37    64    19    69.0   53.3   39.75  3105.69  43.83     78.2
  15. Fritz 17 (Ginkgo) x64           :  3125.09    120    52.9    30    67    23    63.5   55.8   42.28  3106.60  43.77     86.6
  15. Shredder 13 POPCNT x64          :  3125.00    120    52.9    31    65    24    63.5   54.2   42.06  3106.60  43.77     83.8
  17. Fizbo 2.0 BMI2 x64              :  3121.99    120    52.5    30    66    24    63.0   55.0   39.60  3106.67  43.84     86.5
  18. Schooner 2.2 XB SSE x64         :  3115.39    120    51.7    27    70    23    62.0   58.3   39.52  3106.84  43.84     87.8
  19. Andscacs 0.95.123 x64           :  3099.06    120    49.6    25    69    26    59.5   57.5   42.40  3107.25  43.77     87.9
  20. Arasan 22.2 BMI2 x64            :  3090.07    120    48.3    26    64    30    58.0   53.3   41.65  3107.47  43.78     91.2
  -------------------------------------------------------------------------------------------------------------------------------
  20. GullChess 3.0 Sy BMI2 x64       :  3089.71    120    48.3    26    64    30    58.0   53.3   40.57  3107.48  43.81     87.3
  22. Wasp 4.50 Modern x64            :  3086.86    120    47.9    24    67    29    57.5   55.8   41.40  3107.55  43.79     86.2
  23. Combusken 1.4.0 AMD x64         :  3069.96    120    45.8    25    60    35    55.5   50.0   40.77  3107.97  43.81     83.8
  24. Marvin 5.0.0 NN AVX2 x64        :  3063.98    120    45.0    23    62    35    54.0   51.7   41.83  3108.12  43.78     89.6
  24. Winter 0.9 NN BMI2 x64          :  3063.53    120    45.0    24    60    36    54.0   50.0   40.64  3108.14  43.81     85.5
  26. Koivisto 4.19 AVX x64           :  3054.17    120    43.8    24    57    39    52.5   47.5   43.12  3108.37  43.75     86.2
  27. Orion 0.8 NN FMA x64            :  3047.90    120    42.9    20    63    37    51.5   52.5   39.65  3108.53  43.83     89.2
  28. Chiron 4 x64                    :  3045.00    120    42.5    19    64    37    51.0   53.3   39.95  3108.60  43.83     88.4
  29. Halogen 9 NN PEXT x64           :  3038.02    120    41.7    20    60    40    50.0   50.0   40.49  3108.77  43.81     83.1
  30. Nirvanachess 2.5 POPCNT x64     :  3017.82    120    39.2    15    64    41    47.0   53.3   41.82  3109.28  43.78     83.2
  -------------------------------------------------------------------------------------------------------------------------------
  31. Critter 1.6a x64                :  3012.12    120    38.3    19    54    47    46.0   45.0   39.82  3109.42  43.83     85.3
  31. Seer 1.2.1 NN Skylake x64       :  3011.47    120    38.3    16    60    44    46.0   50.0   43.43  3109.44  43.74     82.8
  33. Texel 1.08a13 BMI2 x64          :  3007.70    120    37.9    10    71    39    45.5   59.2   41.79  3109.53  43.78     94.4
  34. Demolito 2020-12-24 PEXT x64    :  3005.24    120    37.5    22    46    52    45.0   38.3   42.24  3109.59  43.77     96.4
  35. Vajolet2 2.8 BMI2 x64           :  3000.88    120    37.1    14    61    45    44.5   50.8   42.72  3109.70  43.76     85.8
  36. Protector 1.9.0 x64             :  2998.14    120    36.7    14    60    46    44.0   50.0   42.94  3109.77  43.75     88.0
  37. chess22k 1.14 JAVA x64          :  2969.44    120    33.3    12    56    52    40.0   46.7   42.11  3110.49  43.77     87.6
  38. Hannibal 1.7 x64                :  2959.54    120    32.1     9    59    52    38.5   49.2   42.91  3110.74  43.75     87.2
  39. iCE 4.0 v853 Modern x64         :  2919.81    120    27.5     9    48    63    33.0   40.0   45.39  3111.73  43.69     87.1
  40. Topple 0.8.0 Modern x64         :  2903.65    120    25.8     6    50    64    31.0   41.7   47.70  3112.13  43.63     83.4
  -------------------------------------------------------------------------------------------------------------------------------
  41. pirarucu 3.3.5 JAVA x64         :  2889.89    120    24.6     7    45    68    29.5   37.5   47.12  3112.48  43.65     84.5
  -------------------------------------------------------------------------------------------------------------------------------

  MoveAvg (average)           : 176.05 / 88.025
  White advantage             : 50.31  +/- 4.97
  Draw rate (equal opponents) : 62.66% +/- 1.24

  Games        :   2.460 (finished)
  White Wins   :     772 (31.4%)
  Black Wins   :     483 (19.6%)
  Draws        :   1.205 (49.0%)
  Unfinished   :       0

  White Perf.  : 55.9%
  Black Perf.  : 44.1%

  ECO A00-A99  :     482 Games (19.6%)
  ECO B00-B99  :     559 Games (22.7%)
  ECO C00-C99  :     545 Games (22.2%)
  ECO D00-D99  :     436 Games (17.7%)
  ECO E00-E99  :     438 Games (17.8%)
I like computer chess!

Post Reply