Feasibility study of compuatational software for radiation damage simulations on GPU based machines
||Feasibility study of compuatational software for radiation damage simulations on GPU based machines|
||NAISS Small Compute|
||Artur Tamm <email@example.com>|
||2023-02-15 – 2024-03-01|
Atomistic simulations of radiation damage studies enable researchers to follow early events taking place in materials being subjected to harsh environmental conditions. This allows to gain insights into events happening very early when a material is being irradiated which is not possible using experimental techniques. The statistics of final products from a collision damage events can be compared with experimental evidence to improve the agreement between the model and measurements. The models can be used to evaluate materials response in a very broad range of conditions without the need to carry out costly experiments for all of them.
The computer codes used in this study range from "ab initio" models based quantum mechanics, such as Qball and INQ real-time TDDFT, to classical atomistic simulations, such as LAMMPS. As the introduction of general purpose GPUs for computational tasks has been very recent, the majority of scientific software has been designed with CPU based clusters in mind. Recently efforts have been made to port packages to GPU systems (for example INQ has been written from scratch with GPUs in mind), but their applicability to specific problems is difficult to assess beforehand.
In this project we aim to test and port two software packages: INQ and LAMMPS with EPH add-on, for radiation damage studies on GPU based machines utilising AMD and NVIDIA products. The INQ software allows for studying the electron-ion coupling from first principles calculations that can be carried over through the EPH model into classical atomistic simulations. LAMMPS is a widely used classical atomistic simulation package which enables to simulate the dynamics in a material consisting of millions of atoms and will be evaluated for radiation damage simulation purposes.
This work will be part of the ANITA project funded by Energimyndiheten, industry and Uppsala university where small modular reactor (SMR) concepts are studied.