Fast, Scalable and Accurate Finite-Element Based Ab Initio Calculations Using Mixed Precision Computing

Electronic structure calculations, especially those using density functional theory (DFT), have been widely employed to predict materials properties across a wide range of materials systems. The role DFT calculations play in the scientific and technological advance is underscored by the fact that ~1/4th of the computational resources for scientific pursuit are utilized for these calculations. Despite the wide applicability in the fields of materials science, chemistry and biology, DFT calculations using state-of-the-art codes have been routinely limited to small system sizes owing to the computational complexity of the problem and the limited scalability of these codes.

In this talk, I will present recent efforts in developing DFT-FE—a massively scalable, accurate and computationally efficient code for fast large-scale DFT calculations. In particular, the talk will focus on developments in the computational framework, algorithmic advances comprising of computational methodologies and mixed precision strategies, and parallel implementation aspects (on many-core and heterogeneous architectures) that have enable fast DFT calculations with unprecedented parallel scalability. The performance metrics of the code, including comparison with other widely used DFT codes, will be presented, along with an outlook to effectively scale to exascale computing platforms.


Download Presentation

Location: Cumberland Amphitheatre Date: August 28, 2019 Time: 11:15 am - 11:45 am Vikram Gavini (U of M)