Since you’re here...

We hope you will consider supporting us today. We need your support to continue to exist, because good entries are more and more work time. Every reader contribution, however big or small, is so valuable. Support "Chess Engines Diary" even a small amount– and it only takes a minute. Thank you.
============================== My email: jotes@go2.pl



Askaig 20260704 - new version chess engine

on
 


Askaig a strong UCI chess engine. Written in C++
Author: Nguyen Van Thang

Askaig is an engine that recently joined our testing program. And it performed well, winning the test match against the Critter 1.6a 8-2! We'll see how the engine develops further.

Askaig 20260704 what's new?

Evaluation — NNUE (replaces the hand-crafted eval)

(768 × 8 king buckets → 512)×2 → 8 material output buckets, SCReLU, horizontal mirroring; default net embedded in the binary (trained on 388M positions), EvalFile to override.
AVX2 / NEON / scalar inference kernels, bit-identical across backends; lazy per-perspective accumulator with finny-table refreshes.
Search — rebuilt

Aspiration windows, singular extensions (+ multicut, double extensions), null move, ProbCut, razoring, reverse futility, LMP, futility, history & SEE pruning, log-formula LMR, 5-table correction history.
Lazy SEE move ordering, TT prefetch after every make, captures-only movegen in quiescence.
Draw handling: repetition, fifty-move, dead material; fifty-move eval damping; Contempt option (default 0).
Time management: node concentration + best-move stability + falling-eval scaling.
Threads

Lazy SMP replaces YBWC (the Split option is gone); reported nodes/nps are summed across all threads. 4 threads search ~3.9× the nodes of 1.
Strength vs 20260628 (1 thread, UHO_4060_v3, SPRT [0, 10] — both H1 accepted)

8s + 0.08s: +373.00 ± 11.73 Elo (6678 games)




Comments

Post a Comment