Releases: UndefinedCpp/EmeraldChessEngine
Emerald Chess 0.4.1
This update does not include significant search or evaluation changes (<10 Elo change)
Updates:
- Support UCI command
go nodes - Shorten PV output after three-fold repetition (and makes
fast-chesshappier) - Move right away if there is an only legal move (enabled automatically only in competition mode)
- Support
Hashoption (default 16, max up to 2048) - Now shows time, nodes per second and hash full information in output
- Added verification search for NMP
- Small tweaks to LMR and move ordering
Emerald Chess 0.4.0
Updates:
- A full rewrite of search algorithms. Fixed bugs and now (finally!) correctly works with history & killer heuristics.
- An update to the
MovePickerclass. Now it won't generate all moves at once, making the engine even faster. - A brand-new NNUE for evaluation. Please find more details in README.
Performance:
Beats previous version with around 91% win rate (i.e. roughly +402 Elo improvement).
Against other opponents:
| Opponent | Version | CCRL Blitz Elo | Elo difference |
|---|---|---|---|
| Neuromancer | 1.5 | 1828 | +173 |
| Cadabra | 2.0.0 | 1960 | +155 |
| Lynx | 0.16.0 | 1970 | +250 |
| Blunder | 5.0.0 | 2014 | +200 |
| Stash | 15.3 | 2170 | +61 |
| Oxide | 1.0.0 | 2238 | -264 |
Estimated Elo: 2100 2211 +/- 20 CCRL blitz scale
All games tested under 10+0.1 time control; played 60 games against each opponent (so the results are not necessarily reliable; the actual outcome may be quite different).
Note: Requires AVX2 and FMA3 support. This should not be a problem for reasonably new CPUs.
Emerald Chess 0.3.0
Updates:
- Static Exchange Evaluation (SEE) implemented
- Late Move Reduction (LMR)
- SEE pruning
- Tuned evaluation and other constants
Performance:
- ~200 Elo improvement compared to version 0.2.0 (Time control 10+0.1)
Note: The binary build requires AVX2 instruction support from CPU. This normally should not cause an issue.
Emerald Chess 0.2.0
Updates:
- Fully refactored search code
- Introduced aspiration window
- A Stockfish-inspired evaluation function
Test result:
- ~147 Elo boost compared to previous version
- ~23 Elo higher than Shallow Blue 2.0.0
Both tested under 10s+0.1s time control.
Emerald Chess 0.1.1
Minor updates:
- Show version message in UCI and on start up
- Insignificant time management update
- Provide binary that can run on machines with AVX512 support (
skylake)
Emerald Chess 0.1.0
The starting point of Emerald chess engine!
Features:
- Principal Variation Search
- Transposition table (not configurable)
- Quiescence search
- Piece square table, mobility, basic pawn structure knowledge in evaluation
- MVV/LVA and killer heuristic in move ordering
- Iterative deepening
Playing strength: estimated ~1600 Elo