Dev Builds » 20260726-1044

You are viewing an old NCM Stockfish dev build test. You may find the most recent dev build tests using Stockfish 15 as the baseline here.

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 14. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN

Commit

Commit ID db98633b1f8bbcad850bee892ec143ff8723ba80
Author Carlos Esparza
Date 2026-07-26 10:44:45 UTC
Avoid recomputing threat/pp accumulation for some king moves. Since the separate threat and psq accumulators were merged in 7c7fe322eae6fa3a594c036eb4fc2553fd4e6a11, any king move requires a full recomputation of the threat (and now also pawn-pair) accumulation, even though these only change when the king switches between the left and right half of the board. This patch adds a method `update_accumulator_hybrid` which incrementally computes the accumulator change of a king move as ``` <new acc> = <previous acc> - <previous psq acc> + <new psq acc> + <threat/pp changes> ``` thus avoiding re-adding all threat/pp features. `<previous psq acc>` and `<new psq acc>` can be computing efficiently using the accumulator cache ("Finny table"). When there are few pieces on the board there typically are only few threat/pp features so computing their sum from scratch is cheaper than computing `<previous psq acc>` from the cache entry. Therefore we only do "hybrid" updates for positions with at least `MIN_PC_COUNT_HYBRID = 15` pieces. For simplicity this also doesn't handle castling moves. passed STC: https://tests.stockfishchess.org/tests/view/6a5b8a7d5529b8472df8158b LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 37792 W: 9855 L: 9557 D: 18380 Ptnml(0-2): 65, 3955, 10578, 4213, 85 local benchmark: ``` Result of 100 runs (counting nodes/1B cycles) ============================================= base (...tockfish.f4bcd4) = 264725 +/- 3700 test (...tockfish.4e77c9) = 266318 +/- 3701 speedup % = +0.60 +/- 0.08 [95% CI; t-test] likelihood(speedup > 0) = 1.0000 CPU: 16 x Intel(R) Core(TM) Ultra 9 185H Hyperthreading: on ``` I don't know if `MIN_PC_COUNT_HYBRID = 15` is optimal. On my computer (avx-vnni) 12 was the optimum, but that did not perform well [on fishtest](https://tests.stockfishchess.org/tests/view/6a5769345529b8472df80d7e). It would be good if other devs could try benchmarking different thresholds and report. The first version of this patch was coded up by gpt-5.5-high. I made many changes, but probably most of the lines of code are LLM-written. closes https://github.com/official-stockfish/Stockfish/pull/6990 No functional change
Copyright 2011–2026 Next Chess Move LLC