Accelerating Parallel Genetic Algorithm for Travelling Salesman Problem Using CUDA Thrust Library

This paper presents a parallel implementation of a genetic algorithm (GA) to solve the Travelling Salesman Problem (TSP) using NVIDIA's Compute Unified Device Architecture (CUDA) SDK on a GPU. The proposed approach exploits the massive parallelism of modern GPUs and the highly optimized primitives of Thrust library to accelerate sorting and elitism selection on GA. The Thrust-based implementation demonstrates significant improvements in computation time and scalability compared to sequential and multi-threaded CPU implementations. Results obtained on multiple TSPLIB datasets show up to 21x speedup with comparable or better solution quality, validating the efficiency of GPU-accelerated metaheuristics implementation using CUDA and Thrust library. © 2026 IEEE.