- c++20
- windows & linux
- uci
- 64-bit
- smp (to 256 threads)
- alpha-beta search
- hash (to 1024 GB)
- ponder
- multiPV
- chess960 (Fischer Random)
- bench, perft & divide
- asychronous cout (acout) class using std::unique_lock std::mutex
- unique NNUE (halfkp_256x2-32-32) evaluation
- visual studio 2022 project files included
- lean & mean codebase size optimizations
- syzygy TBs have been removed (as NNUE evaluation becomes stronger TB's become less valuable)
- MCTS search has been removed
- the source code footprint has been reduced more than 50% from 491 KB to 232 KB
- and the source code has been reformatted via Google style guidelines
- updated binaries available in /src directory (no change to the NNUE file)
- Hash size of the hash table. default is 64 MB.
- Threads number of processor threads to use. default is 1, max = 128.
- MultiPV number of pv's/principal variations (lines of play) to be output. default is 1.
- MoveOverhead Adjust this to compensate for network and GUI latency. This is useful to avoid losses on time.
- Ponder also think during opponent's time. default is false.
- UCI_Chess960 play chess960 (often called FRC or Fischer Random Chess). default is false.
- Clear Hash clear the hash table. delete allocated memory and re-initialize.
- x64 avx2 = fast pgo binary (targeting modern 64-bit systems w/ AVX2 instruction set)
- windows (visual studio) use included project files: fire.vcxproj or fire.sln
- minGW run the included shell script: make_avx2.sh
- ubuntu type 'make profile-build ARCH=x86-64-avx2', etc.
http://www.chessdom.com/fire-the-chess-engine-releases-a-new-version/
- Chess Programming Wiki
- Stockfish
- Ippolit
- Robbolito
- Firebird
- Ivanhoe
- Houdini
- Gull
- nnue-probe
- LittleBlitzer
