DFRC tournament (8 threads)

[Modern Times]

Ongoing DFRC Top 10 (maybe) round-robin tournament.
Each engine running on 8 threads.

Results at the 1,000 games point.
It will probably end at 2,000 games. I don't have enough cores on this machine to use concurrency, so it is slow.

Code: Select all

   # PLAYER               :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  CFS(%)
   1 Reckless 0.9.0       :    3885     28   129.0     200  64.5   68  122   10  61.0    3779      62
   2 Stockfish 18         :    3879     28   127.0     200  63.5   64  126   10  63.0    3780     100
   3 PlentyChess 7.0.0    :    3817     27   108.5     200  54.3   48  121   31  60.5    3787      88
   4 Viridithas 19.0.1    :    3795     25   101.5     200  50.8   32  139   29  69.5    3789      87
   5 Stormphrax 7.0.0     :    3773     26    94.5     200  47.3   30  129   41  64.5    3792      58
   6 Caissa 1.24          :    3769     26    93.5     200  46.8   33  121   46  60.5    3792      51
   7 pawnocchio 1.9       :    3768     27    93.0     200  46.5   28  130   42  65.0    3794      51
   8 Berserk 13           :    3767     25    93.0     200  46.5   35  116   49  58.0    3793      63
   9 Clover 9.1           :    3761     27    91.0     200  45.5   21  140   39  70.0    3793     100
  10 Halogen 16           :    3688     27    69.0     200  34.5   20   98   82  49.0    3802     ---

To date:
Number of Games 1,000
White Percentage 67.05%
White Wins 360 (36.00%)
Black Wins 19 (1.90%)
Draws 621 (62.10%)

The draw rate is higher than on the previous tournament I ran with 1 thread, although engine versions differed.

Tournament Info
DFRC Round-robin tournament of top engines, using randomly selected openings from dfrc_90-120.pgn played reversed sides per pairing
The openings pgn was generated from DFRC_depth20.csv.xz provided I think by the Stockfish team
Thanks to Syn (synthetica9) on Discord for showing me how to extract lines.
108,378 openings fell into the range +90 to +120. I don't recall which Stockfish version generated those evals.

Each engine 8 threads, 1024MB hash, ponder off
Time Control: game in 2 minutes plus 1 second increment, 1'+1"
CuteChess GUI
Hardware: Intel i9 10900 (set at 105W)
Each pairing played reversed sides
Adjudication: Syzygy 5-men only - no eval based adjudication.
Endgame tablebases: Syzygy 5-men.
Ratings calculated with Ordo.

[Nordlandia]

A possible suggestion. DFRC with randomly assigned castling flags. There are 16 permutations for castling flags, worth noting.

[Jouni]

I just played FRC match between SF18 and Reckless. 60 + 0.6 one core:

Code: Select all

Score of Stockfish18 vs Reckless090: 10 - 19 - 171 [0.477]
...      Stockfish18 playing White: 7 - 9 - 84  [0.490] 100
...      Stockfish18 playing Black: 3 - 10 - 87  [0.465] 100
...      White vs Black: 17 - 12 - 171  [0.512] 200
Elo difference: -15.6 +/- 18.2, LOS: 4.7 %, DrawRatio: 85.5 %
200 of 200 games finished.

I was surprised. Stockfish has finally beaten!

[Jouni]

https://www.chess.com/computer-chess-championship is 4 engine final DFRC now.

Which one is correct Time Control: game in 2 minutes plus 1 second increment, 1'+1" ?

[Modern Times]

Oops - mine was Time Control: game in 2 minutes plus 1 second increment, 2'+1"

[Modern Times]

I stopped the tournament at this point:

Code: Select all

   # PLAYER               :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  CFS(%)
   1 Reckless 0.9.0       :    3882     23   196.5     306  64.2  103  187   16  61.1    3778      58
   2 Stockfish 18         :    3879     24   195.0     306  63.7   98  194   14  63.4    3779     100
   3 PlentyChess 7.0.0    :    3812     19   164.0     306  53.6   71  186   49  60.8    3786      82
   4 Viridithas 19.0.1    :    3798     21   157.5     306  51.5   52  211   43  69.0    3788      98
   5 Clover 9.1           :    3767     21   143.0     306  46.7   44  198   64  64.7    3791      61
   6 pawnocchio 1.9       :    3763     20   141.0     306  46.1   41  200   65  65.4    3792      50
   7 Stormphrax 7.0.0     :    3763     20   141.0     306  46.1   42  198   66  64.7    3792      53
   8 Berserk 13           :    3762     21   140.5     306  45.9   48  185   73  60.5    3792      53
   9 Caissa 1.24          :    3761     21   140.0     306  45.8   43  194   69  63.4    3792     100
  10 Halogen 16           :    3699     23   111.5     306  36.4   33  157  116  51.3    3799     ---

Number of Games 1,530
White Percentage 67.16%
White Wins 550 (35.95%)
Black Wins 25 (1.63%)
Draws 955 (62.42%)

Reckless and Stockfish are so close to be inconclusive. Third and fourth would probably not change with more games, and neither would 10th. Fifth to ninth would require a few thousand more games to establish a rock-solid ranking.