GPU acceleration of the Wannier-Fourier interpolation of electron-phonon interactions in EPW
ORAL
Abstract
We present a significant improvement in the computational efficiency of the EPW code (https://epw-code.org) achieved through GPU acceleration of the Wannier-Fourier interpolation of electron-phonon interactions. By leveraging massively parallel computing capabilities of modern GPUs, we demonstrate that a substantial reduction in computation time can be obtained, which will enable the study of more complex systems with higher accuracy. We discuss the implementation details of the GPU acceleration, including the strategies that allowed us to maximize the performance without sacrificing maintainability and portability of the code base. In addition, we will showcase the results of transport calculations, highlighting the improvement in speed and scalability of the new implementation on multi-GPU systems.
*This work is supported by the U.S. Department of Energy (DOE), Office of Science, Basic Energy Sciences (BES) under Award DE-SC0020129. Computational resources were provided by the Texas Advanced Computing Center (TACC) at The University of Texas at Austin.
–
Presenters
-
Tae Yun Kim
- Seoul National University, University of Texas at Austin
- University of Texas at Austin