SC17 Denver, CO

P34: GPU Acceleration for the Impurity Solver in GW+DMFT Packages

Authors: Kwangmin Yu (Brookhaven National Laboratory), Patrick Semon (Brookhaven National Laboratory), Nicholas D'Imperio (Brookhaven National Laboratory)

Abstract: The combination of dynamical mean field theory (DMFT) and GW (or density functional theory) has become a powerful tool to study and predict properties of real materials with strongly correlated electrons, such as high temperature superconductors. At the core of this combined theory lies the solution of a quantum impurity model, and continuous-time quantum Monte Carlo (CT-QMC) has proven an indispensable algorithm in this respect. However, depending on the material, this algorithm is computationally very expensive, and enhancements are crucial for bringing new materials within reach of GW+DMFT. Based on a CPU implementation, GPU acceleration is added, and two times speedup is achieved. New techniques are invented and implemented to deal with various GPU acceleration environment.
Award: Best Poster Finalist (BP): no

Poster: pdf
Two-page extended abstract: pdf

Poster Index