Fine-Grain GPU Parallelization of the Generalized Partition Crossover for Large-Scale Traveling Salesman Problems
Updated
Updated · arxiv.org · Aug 21
Fine-Grain GPU Parallelization of the Generalized Partition Crossover for Large-Scale Traveling Salesman Problems
1 articles · Updated · arxiv.org · Aug 21
Summary
Researchers have developed a fine-grain GPU implementation of the Generalized Partition Crossover (GPX) for large-scale Traveling Salesman Problems (TSP).
The new approach reformulates GPX partitioning as a graph-parallel problem, achieving speedups of up to 625× over traditional CPU methods on instances with up to 2 million cities.
This advancement significantly enhances the scalability of genetic algorithm-based TSP solvers, potentially transforming large-scale combinatorial optimization on modern GPU architectures.