Artificial Neural Network and Non-Dominated Sorting Genetic Algorithm II for the Multi-Objective Optimization of the Graphics Processing Unit Thermal Cooling

This paper proposes an experimental, intelligent optimization approach to improve the thermal cooling performance of an overclocked graphics processing unit (GPU). A closed-loop liquid-cooling system was built and tested utilizing deionized water and a silver (Ag) nanofluid coolant (0.015% vol.) across a variety of microchannel heat sink topologies with varying fin spacing. Key thermal performance indicators, including GPU temperature, coolant outlet temperature, and thermal resistance, were measured at different coolant flow rates. Experiments revealed that raising the flow velocity and decreasing the fin gap considerably enhanced cooling performance, while the Ag nanofluid consistently lowered GPU temperature by 1–3 °C compared to water. An Artificial Neural Network (ANN) surrogate model was constructed and trained using experimental data to support predictive analysis and system optimization, achieving excellent predictive accuracy with low RMSE. The trained ANN model was combined with the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to perform multi-objective optimization, aiming to minimize GPU temperature and thermal resistance while improving heat removal. The Pareto-optimal solutions revealed that nanofluid-based cooling offered the best trade-off circumstances, with optimal designs occurring at moderate flow rates and small fin spacing. The ANN-NSGA-II multi-objective optimization results indicated that the best thermal performance of the GPU cooling system was achieved when using Ag nanofluid (0.015 vol.%) as the coolant, with an optimal coolant flow rate in the range of 1.30–1.84 LPM and an optimal fin/channel spacing of 0.57–0.71 mm, producing GPU temperatures of 29.18–29.66 °C, coolant outlet temperatures of 29.06–29.41 °C, and a minimized thermal resistance of 0.0106–0.0152 °C/W; thus, overall, the suggested ANN-NSGA-II framework works well as a practical design tool for improving GPU cooling systems and may be used to other high-heat-flux electronic thermal management applications.