ML-augmented CFD-framework for heat-sink and fan-speed optimization in gaming workstation

This study presents a system-level CFD-based modeling and optimization of a 328 W gaming workstation's GPU cooling system. A multi-objective Particle Swarm Optimization (MOPSO) algorithm is employed to minimize the maximum GPU temperature (T_GPU) and the total cooling-system cost (C_total). Design variables include longitudinal fin count N_L∈[10,70], transverse fin count N_T∈[10,45], fin shape S_fin∈{Cylindrical,Frustum,Square}, heat-sink material M_hs∈{Al,Cu,SiC}, and fan speed V_fan∈{1000,2000,3000} RPM. A Monte Carlo-based Latin hypercube sampling strategy generated the design matrix, which has 24 design variations for the GPU cooling system. CFD simulations are performed for these 24 combinations, and T_GPU obtained from these solutions is used to train regression models. Out of the various regression models evaluated, artificial neural networks (ANN) based surrogate modeling performs best and accelerates objective evaluations by 70% compared to CFD. The trained ANN predicts T_GPU and C_total for parameter combinations generated via MOPSO. Pareto-optimal solutions illustrate trade-offs between temperature and cost with the optimal configuration of N_L=65, N_T=40, S_fin=Square, M_hs=Cu, V_fan=3000 achieved T_GPU=69.90^∘C and C_total=INR2082.93. Compared to baseline, the optimized design reduced maximum temperature by 24.24%, average temperature by 21.33%, thermal resistance by 34.56%, and improved heat-sink efficiency by 14.10%. Correlation analysis shows V_fan strongly affects T_GPU, while S_fin and M_hs impact C_total. Square fins yield the highest heat flux due to maximal effective surface area and enhanced convective attachment. To our knowledge, this is the first system-level CFD optimization and thermal-flow dynamic study of a gaming workstation. This study additionally introduces crystalline SiC, a high-conductivity, lower-density alternative in electronic cooling, reducing T_GPU by 10 ^∘C over Cu while being 63.73% lighter than Cu and only 18% denser than Al.