Supercomputing (GPGPU) Research
UCL Bartlett School of Architecture
Supervisor, Director of Computing:
Advisor, GAD Programme Leader:
Increasing accessability of supercomputers and computational power is enabling new design methodologies for architects and designers. Simulation is replacing static and parametric modeling as the main medium of design.
This research in particular is focusing on harnessing the extreme potential of parallel computing for massively parallel simulations. Traditionally, CPU s (Central Processing Unit) are responsible for general purpose computation in a computer. They consist of a few, usually 4 or 8, powerful cores. GPU s (Graphics Processing Unit) on the other hand, which are traditionally responsible for no other computation than the graphics display, have tiny and less powerful cores but thousands of them. A GPU becomes much more efficient than a CPU when a parallel computation task is given.
The aim of the research is to exploit the power of GPGPU (General-Purpose Computing on Graphics Processing Units) for simulations that require massive parallelism such as multi agent systems, computational fluid dynamics and other computational physics algorithms. Ongoing experiments yield a performance increase of up to 500x which allows simulation to be used as a real time, intuitive design tool rather than only being a post-design analytical tool.