10,000 Sudoku Puzzles in 76 Milliseconds: Engineering a Fast Human-Strategy Solver

Abstract

HPS is a C++ Sudoku solver that applies 15 named human-style logical strategies without guessing or backtracking. This project studies how to reduce redundant work while preserving that logical approach.

On a selected corpus of 10,000 puzzles solvable by the same 15-strategy system, serial HPS completed the batch in 0.475 seconds, a 35.2× speedup over the serial C++ baseline. With eleven workers on an Apple M3 Pro, it completed the batch in 76 milliseconds, a 220.4× speedup over that same baseline. These results measure performance on supported puzzles, not general solving coverage.