Author: Kieren Pearson Rating Chess Engines Diary CEDR=3547
v.13:
Elo gain compared with Halogen 12:
Elo | 112.92 +- 6.27 (95%)
Conf | 40.0+0.40s Threads=1 Hash=8MB
Games | N: 3436 W: 1177 L: 98 D: 2161
Penta | [0, 40, 720, 797, 161]
http://chess.grantnet.us/test/39622/
Elo gaining patches:
Update default network to 9819aa87.nn by @KierenP in #546
Update default network to 46f749a2.nn by @KierenP in #547
Update default network to 2cc25bae.nn by @KierenP in #548
Update default network to 6913d330.nn by @KierenP in #549
Adjust static evaluation based on 50 move rule by @KierenP in #554
Order loud moves by MVV-LVA by @KierenP in #556
Add capture history heuristic by @KierenP in #563
Tune history values by @KierenP in #567
Add Follow move history heuristic by @KierenP in #574
Add pawn hash indexed history tables for ordering quiet moves by @KierenP in #575
Add quiet move threat history by @KierenP in #576
Allow exact search results to replace lower-bound results even when the score is lower by @KierenP in #593
Static exchange pruning for quiet moves by @KierenP in #594
Improve repetition detection algorithm by @KierenP in #597
Use Cuckoo hash table to detect upcoming repetitions by @KierenP in #598
Add Pawn Correction History by @KierenP in #601
Use partial selection sort to speedup StagedMoveGeneration by @KierenP in #612
Don't evaluate the position when in check by @KierenP in #553
Update default network to 87b3ca62.nn by @KierenP in #616
Use handwritten SIMD inference code (SSE, AVX2, AVX512) by @KierenP in #618
Update default network to b425ef8d.nn by @KierenP in #621
Increase LMR reduction in cut-nodes by @KierenP in #622
Apply negative extensions for non-singular cut nodes by @KierenP in #623
Non pawn correction history by @KierenP in #606
Use fail-medium for RFP fail-high cutoffs by @KierenP in #626
Use non-pawn material scaling to adjust static eval by @KierenP in #628
Use 'improving' heuristic to more optimistically apply RFP by @KierenP in #629
Update default network to 4ed21ecb.nn by @KierenP in #635
Use fail-medium for q-search fail-high cutoffs by @KierenP in #627
Implement node based time management by @KierenP in #638
Make LMP less aggressive on improving nodes by @KierenP in #630
Which binary to choose?
Below are prebuild binaries for Windows and Linux platforms, for a variety of CPU architectures. Please select the most modern target for your architecture.
Target Intel CPU AMD CPU
Legacy Any 64 bit Any 64 bit
SSE4 Nehalem (2008) Bulldozer (2011)
AVX Sandy Bridge (2011) Piledriver (2012)
AVX2 Haswell (2013) Excavator (2015)
AVX2 PEXT Haswell (2013) Zen 3 (2020)
AVX512 Skylake (2017) Zen 4 (2022)
AVX512VNNI Ice Lake (2019) Zen 4 (2022)
Halogen 12.12.3 JA vs other engines:
| Patricia 4.0 | 41.5/68 | +15 | 68 Games |
| Berserk 20241119 | 22/66 | -22 | 66 Games |
| Clover 8.1 | 23.5/62 | -15 | 62 Games |
| Obsidian 14.07 SE | 18/62 | -26 | 62 Games |
| Dark SisTer 8.0 | 3/16 | -10 | 16 Games |
| Stockfish 17 | 4.5/14 | -5 | 14 Games |
| Tucano 12 | 12/12 | +12 | 12 Games |
| Arasan 25.0 | 6/12 | +0 | 12 Games |
| Koivisto 9.2 | 4.5/12 | -3 | 12 Games |
| Big Yuli GM 221024 | 3.5/12 | -5 | 12 Games |
| SF-PRO 04.12.2024 | 3.5/12 | -5 | 12 Games |
| Stockfish 20241122 Ivec | 3/12 | -6 | 12 Games |
| Dark SisTer PB 071124 | 3/12 | -6 | 12 Games |
| SF-PRO2 04.12.2024 | 2.5/12 | -7 | 12 Games |
| Cool Iris 13.90 | 2/12 | -8 | 12 Games |
| Rhetoric 1.4.3 | 9.5/10 | +9 | 10 Games |
| Maestro 1.2.1 JA | 5/10 | +0 | 10 Games |
| Arasan 24.2.2 JA | 4/10 | -2 | 10 Games |
| Yuliana 5.50 | 3.5/10 | -3 | 10 Games |
| Obsidian 14.07 DC | 3/10 | -4 | 10 Games |
| Killfish 021224 | 2.5/10 | -5 | 10 Games |
Our social media:
Comments
Post a Comment