Lecturer
University of York
Phil is a computational physicist and research software engineer at the University of York. He grew up in the 1980s, learning physics at school and computing on his Sinclair ZX Spectrum, before studying Natural Sciences at Pembroke College, Cambridge, followed by a Diploma of Computing Science. He undertook a PhD with Prof. Mike Payne, developing the CASTEP materials modelling program to predict materials chemistry with quantum mechanics; a perfect combination of physics and computing. He has contributed to the development of CASTEP ever since.
First-principles materials modelling on pre-exascale HPC
"First-principles materials modelling" uses quantum mechanics to predict the properties of new and existing materials. These computer simulations have become a cornerstone of materials science, aiding in the interpretation of experimental data and guiding experimental design, and have benefited enormously from a substantial year-on-year increase in available computing power. In recent years, however, the increases in computer power have come about through new kinds of computers, many of which use GPUs to perform the most intensive calculations-Bede is one such machine.Using GPUs effectively requires different techniques to conventional CPUs, and in this project, the CASTEP first-principles materials modelling code will be adapted to use Bede's GPUs much more efficiently. This will dramatically improve the speed at which scientists can perform materials simulations on Bede, in turn leading to more and greater materials discoveries."
Executive summary of project results
We have revolutionised the GPU capabilities of CASTEP, taking it from 1.2X speed-up (pilot project) to around 9.6X speed-up on the same hardware. CASTEP-GPU had a public beta release in 2023, an academic release in 2024, and a full academic and commercial release in 2025.How has your research benefitted from using Bede?
Bede's nvlink enables fast data transfer between the CPU and GPU, which makes a substantial difference to the performance. Bede's hardware and software configuration are well-suited for high-performance computing, and the ability to develop interactively on one of the two login nodes has been invaluable.Has using Bede meant you were able to apply for further research funding?
The PAX-HPC ExCALIBUR project and the CASTEP GPU-eCSE both used Bede performance results to identify bottlenecks, optimise workflows, and support the proposal.Has Bede been a stepping stone for you to access larger HPC facilities?
Yes, the DiRAC Tursa machine and a Director's Discretionary grant for Oak Ridge's former leadership computing facility, Summit. (A similar application to Frontier will be enabled when our current GPU work is complete.)Publications
- Portable Acceleration of Materials Modeling Software: CASTEP, GPUs and OpenACC. Matthew Smith, Arjen Tamerus, and Phil Hasnip. Computing in Science and Engineering 24(1), 46-55 (Jan-Feb 2022) https://doi.org/10.1109/MCSE.2022.3141714