Unlocking Large-Scale Structural Synthesis: High-Performance GPU Topology Optimization For Architectural And Civil Engineering Applications

In this Barcelona symposium talk, Raul Llamas outlines how GPU‑accelerated topology optimization is reshaping large‑scale structural design for both architecture and civil engineering. Drawing on his experience from aerospace wing covers to bridge prototypes, he demonstrates that modern stress‑based methods can handle meshes with hundreds of millions of elements—far beyond the limits of traditional CPU solvers. By focusing on material utilization rather than pure compliance, the approach delivers practical, manufacturable solutions while preserving structural safety across multiple load cases.

Llamas emphasizes the elegance of his implementation: a three‑line core algorithm written in Julia, trimmed to under 50 lines of code, that maps directly onto GPU hardware. A matrix‑free conjugate‑gradient solver equipped with an adaptive preconditioner enables rapid 3‑D iterations; a consumer‑grade 10 GB GPU solved a 26‑million‑element bridge model in roughly two hours, and a high‑end GPU can scale to billions of elements. The workflow reads standard STL, OBJ, and 3MF files, supports region‑specific material properties, and integrates thermal expansion or void definitions, making it versatile for complex civil structures.

Looking ahead, Llamas envisions seamless fluid‑structure interaction and AI‑generated code as the next frontier. By coupling CFD loads with the stress‑driven optimizer, engineers could design façades that react to wind or optimize airfoils in real time. The convergence of AI, GPU hardware, and topology optimization promises a paradigm shift: faster design cycles, richer intermediate solutions, and unprecedented detail for iconic projects like La Sagrada Familia. Companies that adopt these tools now will gain a competitive edge as the industry moves toward fully automated, giga‑scale structural synthesis.