NeuroSpeed is a UCI chess engine that is optimized for extreme time controls. The goal of this chess engine is to see what can be optimized so it can beat other chess engines when it comes to low time, for example, bullet games or ultra-bullet games.
This chess engine is still in development. When I have it in working order and with some results, I will release it as version 1 (v1)
TODO:
- Update the strength table:
| Version | Release Date | CCRL Blitz | CCRL 40/15 |
|---|---|---|---|
| v1.0 | 2024-01-01 | 3000 | 3050 |
| v1.1 | 2024-03-15 | 3100 | 3150 |
| v2.0 | 2024-06-10 | 3200 | 3250 |
These are the features that were implemented so far:
Board Representation:
- Bitboard board representation
- Pre-calculated attack tables
- Magic bitboards
- Hybrid mailbox for piece lookup
- Compact 16-bit move encoding
Move Generation:
- Pseudo-legal move generation
- Separate capture move generation
- All special moves (en passant, castling, promotions)
- Comptime color specialization
Search:
- Negamax with alpha-beta pruning
- Principal Variation Search (PVS)
- Iterative deepening
- Quiescence search
- Principal Variation (PV) table
- Mate distance pruning
Move Ordering:
- PV move ordering
- MVV-LVA (Most Valuable Victim - Least Valuable Attacker)
- Killer move heuristic (2 killers per ply)
- History heuristic
- Incremental move sorting
Evaluation:
- Material counting (incremental)
- Piece-Square Tables (PST) - middlegame/endgame tapered
- Passed pawns
- Isolated pawns
- Doubled pawns
- Pawn chains
- Pawn phalanx
- Piece mobility (knights, bishops, rooks, queens)
- King safety (pawn shelter, attacking pieces)
- Outpost detection
- Bishop pair bonus
- Bad bishop penalty
- Rook on open/semi-open files
- Rook on 7th/2nd rank
- Threat evaluation
- Insufficient material detection
- Special KBN vs K endgame evaluation
- Game phase calculation (opening/middlegame/endgame)
Time Management:
- Fixed depth search
- Fixed time search (movetime)
- Time controls with increment (wtime/btime/winc/binc)
- Simple time allocation
UCI Protocol:
- Full UCI compliance
- Threaded search
- FEN parsing
- Move parsing
- Perft testing
- OpenBench-compatible benchmark
- Huge shout-out to Maksim Korzh for his chess programming series. It's a great help in explaining complex topics in chess engine development.
- Big thanks to the Avalanche chess engine and Lambergar chess engine, where I got a lot of inspiration and a better idea of how to write Zig code, as these chess engines were implemented in Zig.
- Huge shout-out to the Stockfish Discord for being helpful when it comes to questions, and to the Stockfish engine for its great implementation and clean code
- And just like 99% of chess engines, a big thanks to the Chess Programming Wiki for being such a well-structured base of knowledge when it comes to chess engine design