SUPR
EMTO porting to AMD GPU and
Dnr:

NAISS 2023/22-531

Type:

NAISS Small Compute

Principal Investigator:

Wei Li

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2023-05-08

End Date:

2024-06-01

Primary Classification:

10304: Condensed Matter Physics

Allocation

Abstract

EMTO is a density functional theory based tools which have great potential be used for design complex alloy system by both academia and industry. It is parallelized with MPI+openmp. To investigate the 2D and 3D defects in material, eg. screw dislocation and grain boundary, the number of atoms in the simulation cell could be large. For such case, gemm and matrix-inversion of large size double precision complex type matrix will dominate the computing time which could be great benefited from GPU accelerator. Porting EMTO code to GPU with openmp offload on AMD hardware will be also easily transport to GPU from Nvidia and intel later on.