📢 The next SHAREing Flexible Fund call to propose solutions to open tasks is now open until 8 July 2026. View open tasks →
wp1.3

Performance analysis for UK Numerical Relativity community codes

In progress

Fit to programme

This was a proposed solution answering Task 001: Success Story Collection , behind WP 1.1.

Summary

Modern scientific research relies heavily on large-scale simulations running on high-performance computing (HPC) systems. In numerical relativity - a field that studies extreme gravitational phenomena such as black hole and neutron star mergers - multiple research groups across the UK have developed sophisticated simulation codes. However, systematic benchmarking across these codes remains limited, and few provide examples designed for performance analysis.

This task will establish a shared benchmarking framework for UK numerical relativity codes. The framework will provide a standardised test problem and openly available performance measurements that can be applied across different HPC systems and testbeds. This will enable researchers to compare computational performance, understand scaling behaviour, and identify optimisation opportunities.

The task will also use numerical relativity codes as realistic scientific applications to test and refine the performance analysis practices outlined in the SHAREing Performance Analysis Workbook, providing practical case studies that can be used to identify any gaps or sections that need clarifying.

All benchmarking workflows, documentation, and results will be made openly available through our UKNR public repository, and in a short “success story” report on the SHAREing website, supporting stronger Digital Research Infrastructure across the UK.

Outputs

  • Definition of a model test problem for NR codes, and publicly available setups/workflows for benchmarks for main UK numerical relativity codes – e.g. BAM, ETK, ExaGRyPE, GRChombo, GRTeclyn, MHDuet. Documentation on how to run on a typical system, and how to extend these for other codes or testbeds. These will be stored on the UKNR github organisational pages: (see https://github.com/UKNumericalRelativity/uknr-benchmarking)

  • Report on current performance status of range of codes on typical GPU and CPU systems -report to be made public on UKNR website and disseminated at UKNR events.

  • At least one publicly available example demonstrating the application of the SHAREing Performance Analysis Workbook methodology to a numerical relativity code or benchmark.

  • A short report or summary suitable for publication on the SHAREing website describing our “success story” - the benchmarking framework, results obtained, and lessons learned for the wider DRI community.