Releases: jeffreyan11/laser-chess-engine
Laser 1.7
Roughly 3250 ELO (CCRL 40/4)
Regressions vs. 1.6:
+114 in 12000 games at 3+0.02
+127 in 12000 games at 15+0.05
Major changes from 1.6:
- Search and evaluation tweaks
- Speed improvements and bugfixes, especially on multiple threads
- Remove EvalCache option
Four binaries are provided:
- Laser-1_7-BMI2 for Haswell or later processors with BMI2 instruction support
- Laser-1_7 for 64-bit machines with POPCOUNT instruction support
- Laser-1_7-NOPOPCNT for 64-bit machines without POPCOUNT
- Laser-1_7-ARMv8 for 64-bit Android phones
Laser 1.6
Roughly 3180 ELO (CCRL 40/4)
Regressions vs. 1.5:
+151 in 12000 games at 3+0.02
+151 in 12000 games at 15+0.05
Major changes from 1.5:
- Search and evaluation tweaks
- SMP rewrite
- Time management rewrite
Three binaries are provided:
- Laser-1_6-BMI2 for Haswell or later processors with BMI2 instruction support
- Laser-1_6 for 64-bit machines with POPCOUNT instruction support
- Laser-1_6-NOPOPCNT for 64-bit machines without POPCOUNT
Laser 1.5
Roughly 3110 ELO (CCRL 40/4)
Regressions vs. 1.4:
+84 in 8000 games at 3+0.02
+97 in 4000 games at 15+0.05
Major changes from 1.4:
- Multiple bugfixes
- Low depth pruning changes
- Evaluation tuning
- Implement pondering
- This is a preliminary release of pondering, which may not work with GUIs other than Cutechess. Pondering support will continue to improve in future versions.
Three binaries are provided:
- Laser-1_5-BMI2 for Haswell or later processors with BMI2 instruction support
- Laser-1_5 for 64-bit machines with POPCOUNT instruction support
- Laser-1_5-NOPOPCNT for 64-bit machines without POPCOUNT
Laser 1.4
Roughly 3020 ELO (CCRL 40/4)
Regressions vs. 1.3:
+86 in 8000 games at 3+0.02
+108 in 4000 games at 15+0.05
Major changes from 1.3:
- Multiple bugfixes (Syzygy, UCI parsing, eval)
- Add countermoves history
- Evaluation tuning
- Lots of new evaluation terms
- New UCI options for scaling material and king safety value
Three binaries are provided:
- Laser-1_4-BMI2 for Haswell or later processors with BMI2 instruction support
- Laser-1_4 for 64-bit machines with POPCOUNT instruction support
- Laser-1_4-NOPOPCNT for 64-bit machines without POPCOUNT
Laser 1.3
Roughly 2950 ELO (CCRL)
Regressions vs. 1.2:
+131 in 8000 games at 3+0.02
+128 in 4000 games at 15+0.05
Major changes from 1.2:
Syzygy TB support
Evaluation tuning using a method based on Texel's Tuning Method
LMR at PV nodes
Change LMR formula, create a search stack to provide more information for pruning decisions (ideas from Stockfish)
Three binaries are now provided.
- Laser-1_3 for 64-bit machines with POPCOUNT support
- Laser-1_3-NOPOPCNT for 64-bit machines without POPCOUNT
- Laser-1_3-x86 for 32-bit machines. The 32-bit compile is not optimized very well - this may improve in a future release.
Thanks also to Michael Byrne for providing a Mac compile: https://www.dropbox.com/s/w1juxbhxpzjyr84/uci-chess-engine-1.3.zip
Laser 1.2
Roughly 2850 ELO (CCRL)
Regressions vs. 1.1:
+114 in 8000 games at 3+0.02
+109 in 4000 games at 15+0.05
Major changes from 1.1:
Improved king safety
Improved passed pawn evaluation
Slight changes in pruning and reduction
Faster compile
Bugfixes
Provided binaries are for 64-bit Windows. Use the NOPOPCNT version for Intel processors before 2008 or non-Intel processors. Binaries have only been tested on 64-bit Windows and Intel processors.
See README for more information.
Laser 1.1
Stable release
Roughly 2780 ELO (CCRL)
Regressions vs. 1.0:
+60 in 8000 games at 3+0.02
+68 in 4000 games at 15+0.05
Major changes from 1.0:
Fail-soft and aspiration windows
Pawn-related tuning
Added king-pawn tropism
New LMR formula
Provided binaries are for 64-bit Windows. Use the NOPOPCNT version for Intel processors before 2008 or non-Intel processors. Binaries have only been tested on 64-bit Windows and Intel processors.
See README for more information.
Laser 1.0
Stable release
Roughly 2700 ELO (CCRL)
Regressions vs. 0.2.1:
+97 in 10000 games at 3+0.02
+108 in 2000 games at 15+0.05
Major changes from 0.2.1:
Bugfixes (prevent connection stalls and fix crashes)
Lazy SMP up to 32 threads (tested on up to 4)
Better three-fold detection
Improved eval (rooks on open files, knight outposts, mobility)
Provided binaries are for 64-bit Windows. Use the NOPOPCNT version for Intel processors before 2008 or non-Intel processors. Binaries have only been tested on 64-bit Windows and Intel processors.
See README for more information.
Thanks to Michael Byrne for providing a Mac compile: http://talkchess.com/forum/viewtopic.php?topic_view=threads&p=653416&t=58680
Laser 0.2.1
Slightly more stable than 0.2.
Laser 0.2
Semi-stable release: crashes occasionally in near-mate positions (once every few thousand games)
Update 10/29: crashes appear to occur more frequently than expected. This will be patched as soon as possible.
Roughly 2600 ELO
Regressions vs. 0.1:
+201 in 5000 games at 3+0.02
+250 in 2000 games at 15+0.05
Major changes from 0.1:
Bugfixes (LMR and hashing, UCI timing parsing)
Better scaling
Extensions and more pruning (futility and late move)
Improved eval (passed pawns, king safety, basic endgames)
Provided binaries are for 64-bit Windows. Use the NOPOPCNT version for Intel processors before 2008 or non-Intel processors. Binaries have only been tested on 64-bit Windows and Intel processors.
See README for more information.