Massively Multicore Parallelization of Kohn-Sham Theory.

Area of Science:

• Computational Chemistry
• Materials Science
• High-Performance Computing

Background:

• Kohn-Sham density functional theory (KS-DFT) is a cornerstone of modern computational materials science.
• Scaling KS-DFT calculations to larger systems is crucial for advancing scientific discovery.
• Accelerator technologies offer potential for significant performance gains in scientific computing.

Purpose of the Study:

• To describe a novel multicore parallelization strategy for KS-DFT.
• To leverage ClearSpeed Technology accelerator hardware for enhanced computational performance.
• To achieve efficient parallelization of KS-DFT on a large-scale computing cluster.

Main Methods:

• Implementation of a multicore parallelization scheme for KS-DFT.
• Utilization of ClearSpeed Technology accelerator cards.
• Reformulation of the Coulomb problem using Poisson density fitting.
• Application of numerical quadrature for three-index integrals.

Main Results:

• Successful parallelization of KS-DFT across 2304 cores.
• Demonstration of efficient scaling on the parallel architecture.
• Validation of the reformulated Coulomb problem approach with negligible errors.
• Significant speedup achieved using the accelerator technology.

Conclusions:

• The described parallelization method enables efficient large-scale KS-DFT computations.
• ClearSpeed Technology accelerators provide a viable pathway for accelerating KS-DFT.
• The approximations introduced are validated and do not compromise accuracy.
• This work paves the way for tackling more complex materials science problems.