PlentyChess - A UCI chess engine with a neural network based evaluation.
Rating Chess Engines Diary CEDR=3661
Building on Windows
For the build to work on Windows, you need to install MinGW as well as the required packages (make, g++, pthread, etc.).
Then you can use the makefile as you would on a unix system.
v.2.0.0:
Performance vs. PlentyChess 1.0.0
STC
Elo | 61.20 +- 3.49 (95%)
Conf | 8.0+0.08s Threads=1 Hash=16MB
Games | N: 20000 W: 6995 L: 3508 D: 9497
Penta | [40, 1356, 4318, 3649, 637]
https://chess.aronpetkovski.com/test/567/
LTC
Elo | 49.14 +- 4.76 (95%)
Conf | 40.0+0.40s Threads=1 Hash=64MB
Games | N: 10000 W: 3144 L: 1739 D: 5117
Penta | [2, 744, 2253, 1849, 152]
https://chess.aronpetkovski.com/test/568/
Improvements
This version includes the following improvements compared to version 1.0.0:
Multithreading improvements due to thread voting
Much stronger NNUE architecture featuring king buckets
Many small improvements to the search algorithm and move selection
UCI "Ponder" option
Various bugfixes, including ones that were affecting strength
Binary guide
generic: Slowest
avx2: Faster, compatible with most modern CPUs
bmi2: Faster, includes avx2, not recommended for Zen1/2 CPUs
avx512: Fastest, includes bmi2
A big thank you to the members of my new OpenBench testing instance for our great collaboration.
Comments
Post a Comment