Architecture / Engineering / Construction
Arup Cuts Structural Simulation From Hours to Minutes With NVIDIA DGX Spark
Objective
Arup is a world-leading built environment consultancy delivering engineering, design, and advisory services across more than 130 countries. As the industry increasingly prioritizes reducing embodied carbon while maintaining structural performance, Arup’s engineers must accelerate the design-to-simulation cycle—without extending project timelines. Traditional finite element method (FEM) simulations are computationally intensive, limiting the early-stage iteration needed to identify optimal, low-carbon designs.
By training and deploying graph neural network (GNN)-based surrogate models on NVIDIA DGX Spark™, Arup reduced structural simulation evaluation time from hours to minutes while achieving up to 95% accuracy—enabling faster, data-driven design decisions at scale.
Customer
Arup
Partner
Tsinghua University
Topic
Simulation / Modeling / Design
Key Takeaways
Hours to Minutes
- GNN surrogate models reduced the time required for structural performance from hours to minutes, enabling near-real-time design iteration in early-stage workflows.
~90 to 95% Accuracy
- GNN surrogate models achieved up to 95% prediction accuracy compared to traditional FEM simulations, maintaining engineering-grade fidelity.
Up to 50% Faster
- DGX Spark delivered up to 50% faster GNN model execution and training compared to alternative platforms, accelerating development cycles.
Scaling Structural Design Exploration
Arup sits at the intersection of engineering ambition and environmental responsibility, as a global consultancy trusted to deliver some of the world's most complex structures—from supertall skyscrapers to offshore wind farms, across more than 130 countries and every major market. Sustainability is not a constraint for Arup; it is the brief. That means every structural design decision carries a dual mandate: perform to specification, and use as little material—and generate as little embodied carbon—as possible. Finding the design that satisfies both demands requires exploring thousands of configurations. And exploring thousands of configurations requires simulation.
In structural engineering, the finite element method (FEM) is the industry standard for simulating structural behavior and ensuring safety. However, these simulations are computationally intensive—often taking minutes to hours per model and requiring HPC infrastructure and associated queues—limiting the number of design-to-simulation cycles that can be evaluated in project timelines. This is particularly impactful in early-stage workflows where rapid iteration is critical. As projects increasingly require balancing structural performance, material cost, and embodied carbon, this constraint becomes a significant bottleneck.
Engineers are often forced to rely on simplified assumptions or evaluate only a limited set of options, reducing the ability to identify the best-performing, most sustainable solutions. For a global consultancy like Arup—delivering complex projects across more than 130 countries—this limitation impacts both efficiency and innovation, making it difficult to scale design exploration and deliver low-carbon, optimal outcomes within competitive project timelines. What Arup needed was a way to explore thousands of structural options in real time—without sacrificing engineering rigor or extending project timelines.
Transforming Early-Stage Design With Quantitative AI Insights
By deploying GNN-based surrogate models on NVIDIA DGX Spark, Arup transformed its early-stage structural design workflow. Structural performance evaluation time dropped from hundreds of hours to just minutes, enabling near-instant predictions during design studies. The models achieved 90–95% accuracy across key structural metrics, ensuring engineers can confidently use AI-driven insights for critical design decisions.
In early-stage design, where numerous parameters generate many potential schemes, engineers are often limited to qualitative assessments due to the time required for analysis. In a recent study, Arup generated multiple building massing options through parametric modeling, each requiring rapid structural evaluation to inform early design decisions. With GNN-based predictions running on DGX Spark, thousands of structural schemes can now be evaluated efficiently—enabling more informed trade-offs across structural performance, cost, and embodied carbon.
Running locally, DGX Spark maintains secure data handling while supporting faster project delivery and more sustainable design outcomes. Engineers who previously could only perform qualitative assessments during early design phases can now leverage quantitative, data-driven insights to identify optimal, low-carbon solutions faster.
“Arup’s advanced AI, developed in partnership with Tsinghua University, combined with NVIDIA DGX Spark and AI optimization expertise, is enabling us to design complex structures faster, reducing both cost and carbon, while keeping project data local and secure.”
Will Cavendish,
Global Digital Services Leader, Arup
Looking Ahead
Physical simulation is central to Arup's highest-value work, spanning solids, fluids, and gases across applications such as wind modeling, pollution and fire propagation, hydrological studies, and airborne infection analysis. Two methodological advances are currently driving this field forward: agentic AI, which orchestrates complex engineering workflows more efficiently, and surrogate models, which replace classical equations with neural networks to achieve dramatically faster computation without sacrificing accuracy.
Discover how NVIDIA DGX Spark is accelerating AI-powered engineering simulation and sustainable design.