by Cris CeckaInstitute for Applied Computational Science, Harvard University
I will present some of the key choices and motivations for developing finite element assembly methods on the GPU. The application of these methods yield a real-time demonstration of a nonlinear hyperelastic solid with the assembly, solution, update, and visualization stages performed solely on the GPU. Additionally, this talk will discuss a novel sparse matrix structure and simple variations of existing structures that accelerate common matrix computations.
Cris Cecka received his Ph.D. in Computational Mathematics from Stanford University in 2011. He now now works as a Lecturer/Researcher at Harvard University in the new Institute for Applied Computational Science. His research focuses on finite element methods, boundary element methods, and high performance computing. You can contact him at email@example.com.