Since you’re here...

We hope you will consider supporting us today. We need your support to continue to exist, because good entries are more and more work time. Every reader contribution, however big or small, is so valuable. Support "Chess Engines Diary" even a small amount– and it only takes a minute. Thank you.
============================== My email: jotes@go2.pl



Arcanum 2.8 - new version chess engine



Arcanum is a free UCI chess engine under the MIT license. 
Author:Lars Aurud, Rating Chess Engines Diary CEDR=3321

Arcanum 2.8 - what's new?

Performance Changes
Add pawn correction
Improve NNUE quantization in feature transformer a accumulator from 127 to 255
In PV, avoid pruning moves based on history scores.
Include continuation history when pruning moves based on history.
Increase initial aspiration window size from 25 to 35
Increase base reduction in Null Move Pruning (NMP) from 2 to 4
Remove Internal Iterative Reductions (IIR). It did not affect performance, and was thus removed to simplify the code.
Allow Late Move Reduction (LMR) of checking moves
Filter moves in ProbCut using Static Exchange Evaluation (SEE) before searching
Improve TT pruning by fixing bound check for upper bound entries (using <= instead of <)
Increase the baseline reduction in Late Move Reduction (LMR), and adjust it based on history score.
Bugfixes and Other Improvements
Require 2 repetitions in pre-root positions to consider it a repeat in search. Thus a position is not automatically evaluated as a draw if it has occurred earlier in the game. Still only required a single repeat in the search tree.
Force search to always complete the first iteration
Ensure enpassant only affects the hash if the enpassant is legal.
Fix a bug where the PV-line could contain an invalid sequence of moves
Fix a bug where a leaf node with could be evaluated as a stalemate because hasMove() only checked the first piece of each type.
Improve FEN parsing performance by ~4x
Elo Estimate
Arcanum v2.8 is estimated to be 51-64 Elo stronger than Arcanum v2.7 depending on time control.

Self Play
Opening Book: UHO_Lichess_4852_v1.epd
Tablebase: 5-man syzygy

Time control: 120s + 1s Hash: 128MB
Results of v2.8 vs v2.7
Elo: 63.97 +/- 13.89
LOS: 100.00 %, DrawRatio: 44.22 %, PairsRatio: 3.83
Games: 692, Wins: 241, Losses: 115, Draws: 336, Points: 409.0 (59.10 %)
Ptnml(0-2): [0, 40, 153, 140, 13], WL/DD Ratio: 0.96

Time control: 10s + 0.1s Hash: 32MB
Results of v2.8 vs v2.7
Elo: 51.16 +/- 13.47
LOS: 100.00 %, DrawRatio: 41.10 %, PairsRatio: 2.56
Games: 944, Wins: 317, Losses: 179, Draws: 448, Points: 541.0 (57.31 %)
Ptnml(0-2): [8, 70, 194, 176, 24], WL/DD Ratio: 0.92

Arcanum 2.7 vs other engines:

Artemis 18.7TR1/4-24 games
ShashChess 401/4-24 games
Kookaburra 3.101/4-24 games
Raid 4.71/4-24 games
Caissa 1.230.5/4-34 games
Dark SisTer v60.5/4-34 games
Obsidian 16.15 SE0.5/4-34 games
Omega Explorer Oct250.5/4-34 games
Stockfish dev-202510050.5/4-34 games
HypnoS 1.000.5/4-34 games
JigSaw 6.00/4-44 games
Steel Fighter III0/4-44 games
ShashChess 39.1 JA0/4-44 games
CorChess 202510050/4-44 games
BlackFish 2705250/4-44 games
Sapling 1.2.61.5/3+03 games
Prune 2.0.12/2+22 games
Tcheran 6.02/2+22 games
Reyna 1.02/2+22 games
Colossus 2025b2/2+22 games
RukChess 4.2.0 JA2/2+22 games
 

github:https://github.com/LarsAur/Arcanum/releases/

#chessenginesdiary
#Arcanum

Comments