Sublime Design: How GPU Computing Is Changing the Shape of Architecture

by Bea Longworth

Combine the Concorde airplane’s swooping lines, the Sydney Opera House’s soaring spaces and the intricacies of a bird’s nest, and you’ll begin to describe the transformative work of architect Daghan Cam.

Cam — a teaching fellow at University College London’s Bartlett School of Architecture, whose resume includes time working with Zaha Hadid, the Pritzker Prize winning architect — is using GPU computing to render stunning, abstract 3D-printed designs and train image-processing robots to construct complex structures.

Work of Daghan Cam
Architecture as art: UCL teaching fellow Daghan Cam’s 3D printed prototype, on show.

Cam, who did his initial training in Istanbul before coming to London six years ago, uses computational physics and myriad algorithms to simulate the structures of materials.

He’s been coding with the CUDA parallel programming model and using NVIDIA GPUs for years. That includes his latest project exploring the capabilities of large-scale 3D printing combined with deep learning for robotic fabrication.

Cam designed a 3D model using a Quadro K6000 graphics card and Tesla K40 GPU accelerator. Then he worked with Boston Limited and Materialise to print his high-resolution simulation using a mammoth stereolithography 3D printer, which creates large, complex components in a single piece.

The prototype is abstract — and beautiful. It’s the kind of work you’d find in a museum of contemporary art. But the project has a practical focus: using minimal materials to reduce building costs, without compromising structural integrity.

Detail of work by architect Daghan Cam
Cam’s latest works focuses on large-scale 3D printing and robotic fabrication processes.

The prototype was a hit at Milan Design Week 2015 this spring. Now, the Bartlett School of Architecture is backing Cam’s work on large-scale 3D printing and robotic fabrication. He and his team are exploring computer vision and deep learning algorithms for real-time image processing and feedback during robotic fabrication processes.

This new stage of the project aims to develop advanced robotic construction techniques to build complex structures efficiently. They’ll take advantage of autonomous, decision-making robotic behavior by training industrial manufacturing robots using the NVIDIA cuDNN deep neural networks library. NVIDIA GPUs will process the image data collected by sensors on the robots.

The results, and the prototype, were on display at UCL’s B-Pro Show earlier this month. The video below shows some of simulations Cam performed.