The benefits of an EPCC internship

16 April 2024

Thomas Flynn, a final-year PhD student in Applied Maths at Newcastle University, has recently completed a three-month internship with EPCC. Here he writes about the experience.

Thomas Flynn

I first became interested in high performance computing (HPC) through my research, which uses computational simulations to study ultracold quantum gases. One of the main drivers behind my growing interest in the subject was the completion of several ARCHER2 training courses. Following these courses I enquired about an internship at EPCC, as it is the leading centre for HPC in the UK, and was pleased to be accepted.

Benchmarking GPUs

During my time at EPCC I worked alongside EPCC's Mark Bull and James Richings, and fellow EPCC intern Chris Rae, benchmarking a variety of GPUs. As HPC moves into the Exascale age, GPUs will begin to dominate HPC systems and thus centres such as EPCC must diversify to meet this growing need. 

GPUs are notorious for their variety in programming models, eg CUDA/HIP, OpenACC and many more. My main focus was to collate a selection of benchmarks which support OpenMP-offloading to GPUs and then build up a dataset of performance metrics to compare OpenMP with the native GPU programming languages CUDA and HIP. I worked across five GPU platforms: 

  • AMD Instinct MI210 on ARCHER2 (an EPSRC Tier-1 national supercomputer system)
  • NVIDIA V100 on Cirrus (an EPSRC Tier-2 system)
  • NVIDIA A100 and H100 on the Edinburgh International Data Facility (EIDF)
  • NVIDIA GH200 on the state-of-the-art Grace-Hopper integrated CPU+GPU memory chip.

Working with such a variety of systems has been a challenge as I’ve worked across different compilers, profilers and hardware architectures, with plenty of debugging. However, one of the most significant challenges for me personally was working with the EIDF. As stated above, I have worked with traditional HPC systems throughout my PhD research, whereas the EIDF is a containerised system. Therefore, I had to familiarise myself with a list of technologies new to me, such as containerisation with Docker to build reproducible benchmarks, and Kubernetes to deploy my benchmarks on the EIDF. As a result, I am now able to work in cloud-like environments.

My time at EPCC has been an enormous privilege. I have gained experience crucial to my career development by working with a diverse selection of computing facilities, and have been immersed in a working environment that has allowed me to learn so much from my colleagues. 
 

Links

ARCHER2 training
https://www.archer2.ac.uk/training

Author

Thomas Flynn, Newcastle University
t.flynn@ncl.ac.uk