Powerful C++ Chess Engine - to conquer the chess board

Overview

BlackCore is a neural network based chess engine developed from scratch. Its alpha beta search uses various pruning techniques, powered by blazing fast move generator.

Playing strength - Last updated: 2023. 2. 12.

Version	CCRL 2'+1" elo	CCRL 40/15 elo	CEGT 40/4 elo	SPCC 3'+1"
v5.1		3281	3242
v5.0		3167	3143	3251
v4.0	3182	3135	3068
v3.0	3069	3035	2941
v2.0		2982
v1.0	2134

Installation

Downloading prebuilt binary

You can download the latest release here both for Windows and Linux. To select the right binary use the first instruction set that your CPU supports (doesn't crash), in the order of BMI2 -> AVX2 -> popcnt

Building from source (recommended)

After downloading the source, you can run the following commands, to build a native binary. This option gives the best performance. Please update your compiler before building!

With any questions or problems feel free to create a github issue.

cd src
make clean build ARCH=native

ARCH = popcnt/avx2/bmi2/native

If you wish to use another compiler than g++ set the CXX variable to for example clang. Warning: compatibility is not guaranteed.

Usage

BlackCore in itself is a command line program, and requires a UCI compatible Chess GUI (like Cute Chess or Arena) for the best user experience.

UCI Options

• Hash - The size of the Hash table in MB.
• Threads - The amount of threads that can be used in the search
• Move Overhead - The delay (in ms) between finding the best move and the GUI reacting to it. You may want to make this higher if you notice that the engine often runs out of time.
• SyzygyPath (Optional) - The folder containing Syzygy tablebases.
• EvalFile (Optional) - The file containing the neural network which should be use. If it isn't found BlackCore will use the embedded network.

Files

This project contains the following files:

• README.md the file that you are reading.
• LICENSE containing the license of this repository.
• .github folder contains automated GitHub workflows like building this project.
• src folder contains the source code of BlackCore
• scripts folder contains short scripts

Features

• Hopefully an easy to read and well commented source
• UCI support
• Perft test
  • Up to ~240M nps (with NNUE accumulator disabled)
    • Intel i3-7100 3.9Ghz CPU
    • Single-threaded
    • Hashing disabled
• Benchmark
  • Fixed depth search on a set of custom positions
• 16 bit encoded moves
• Bitboard representation
• Engine
  • Search
    • Parameters tuned using weather factory
    • Iterative deepening
    • Aspiration window
    • Alpha-Beta
      • Negamax
      • Transposition table
        • Entry aging
        • Bucket system
      • Principal variation search
        • Late move reduction/extension
          • R = max(2, LMR_BASE + (ln(moveIndex) * ln(depth) / LMR_SCALE)));
        • Move count/late move pruning
        • Futility pruning
        • Singular extension
        • Check extension
      • Razoring
      • Reverse futility pruning
      • Null move pruning
    • Quintessence search
      • Stand-pat
      • Delta pruning
      • Static-exchange-evaluation pruning
    • Move ordering
      • Hash move
      • SEE
      • Killer, counter and history heuristics
    • Multithreading support
      • Lazy SMP
  • Time management based on search stability

NNUE Ethics

As of v2.0 BlackCore uses neural networks for the evaluation of positions with support for AVX2 vectorization. There is no need to worry about installing networks separately as they are embedded using incbin. Every net was trained by me with training code which I wrote. Unfortunately I lack the hardware resources which are needed to generate data, which means that external engines are used for that purpose. In my opinion this requires transparency about the origins of the training data and acknowledgement of these engines.

Berserk by Jay Honnold

Berserk is very strong chess engine that generated the training data with the contribution of Shaheryar Sohail which were used continuously throughout v2.0-v5.0.

Leela Chess Zero by The LC0 Team

LC0 uses a different approach for playing high level chess. It's powered by MCTS and acquired all of her chess knowledge by selfplay. Since v6.0-dev BlackCore uses data from here, which is licensed under Open Database License.

Also thanks to...

Zuppa Cipolle

For teaching me SPRT and letting me use his OB instance.

Smallbrain by Disservin

Smallbrain is an awesome engine that helped me understand many important concepts, and a very special thanks to Disservin for giving me many great ideas how can I further improve BlackCore.

Chess Programming Wiki

The Chess Programming Wiki is the greatest resource for everybody who wants to be informed about the basics and the state-of-the-art technologies of chess programming.

OpenBench by Andrew Grant

OpenBench is a SPRT testing framework, used for the testing of different techniques in BlackCore

Shaheryar Sohail

Sohail (developer of StockNemo) guided me through many problems regarding NNUE and without .

Weather factory by Pali

Weather factory was used to tune various parameters of BlackCore using the SPSA method.

Stockfish by The Stockfish team

Thanks to the Stockfish team for making such a wonderful and an easy-to-read codebase, that inspired me to get into chess programming in the first place.