Lead: Martyn Guest and Jose Javier Munoz Criollo (Cardiff University)

Description

This miniproject looks to showcase the performance of established community GPU-enabled codes extensively used in life sciences and genomics. The codes chosen – AMBER, GROMACS, LAMMPS and NAMD feature heavily on both local and regional clusters. AlphaFold AI (Dorado) will also be included in the mix given its widespread adoption and GPU utilisation by the Genomics community.

The objective is to capture the codes’ performance as a function of the code implementation (CPU-only or GPU accelerated), the number and type of GPUs used, as well as the level of utilisation of key GPU features (memory, CUDA cores, Tensor cores, interconnect, compute capability, etc). Thus, the aims of this mini-project are to:

  • demonstrate the performance improvements of the selected codes on GPUs compared, for example, to those seen on CPU systems.
  • Measure the impact on performance related to the availability of various GPU features.
  • provide a rapid demonstration of SHAREing’s presence and capabilities to the RTP community, many of whom will be supporting these codes on their local clusters. Outputs will include a technical report / white paper on the findings, to be published on the SHAREing web, along with presentations at community events and/or workshops (e.g. CIUK’25).

Outcomes

This project is still running, and outcomes will be published here once the miniproject has finished.