Fit to programme

This was a proposed solution answering Task 041: Develop training on green software practices for accelerate computing, behind WP 2.3.

Summary

In response to the impending global climate crisis, the UK Government has committed to achieving net zero carbon emissions by 2050, with several sectors required to reach this by 2040. One vital and growing sector is that of largescale computing and data facilities, with data centres alone estimated to contribute over 0.5% of global annual CO2 emissions. This expansion is driven by both direct numerical simulations and the explosion of AI/ML methodologies, and the widespread adoption of GPU-accelerated architectures. It is thus important to ensure that there is training for software developers and data science researchers on the best “green computing” practices. Although there are some training courses available for “green” best practices in High-Performance Computing (HPC), none of these focus specifically on best practices with GPU-based accelerators, despite their rapid adoption in HPC from largescale data centres to workstations and laptops. We will create a training course that focuses particularly on green GPU usage that covers the impact on climate of: the manufacture and use of accelerator hardware; the relationship between optimisation and carbon emission, and how this compares to non-accelerated HPC; and overall best practices for using GPUs in a climate conscious way.