The Natural Environment Research Council (NERC) Earth Observation Data Acquisition and Analysis Service (NEODAAS) is based at Plymouth Marine Laboratory (PML) and overseen by the National Centre for Earth Observation (NCEO) on behalf of NERC. NEODAAS offer a range of services to assist researchers in their use of Earth Observation data including; acquisition and processing of data from satellites, development of new products, use of AI algorithms and user support.
Digital research infrastructure has been highlighted as a key focus for improvements in sustainable research and computing. In a recent study, colleagues at NEODAAS took existing code being used by researchers in NCEO and investigated the advantages of using different software libraries to translate the code to run on GPUs. The study highlighted what type of code is best suited to for accelerating using GPUs.
Comparisons were made between the time taken to run the code on the original Central Processing Unit (CPU) and the MAGEO CPU and GPU. The results were substantial; accelerations of between x30 and x1800 were achieved in code processing which in terms of energy savings is estimated between a 93.9% and 99.8% reduction in electricity usage.
Though only a small scoping study, this project has been successful in demonstrating the benefits of using GPUs to accelerate code, and the importance of research software engineers in advising on how to improve existing code, but this has just scratched the surface of what is possible. Suggestions for future work in this area have been presented, which could lead to significant performance improvements of NCEO code using GPUs. As NCEO process increasingly larger volumes of data improving the efficiency of code used is an important step as part of moving towards UKRI’s NetZero aim for Digital Research Infrastructure.
Thanks to the success of the study, additional funding from the (NERC), has been awarded to NEODAAS so they are able to work with more researchers to optimise their code.
Researchers interested in working with NEODAAS to optimise their code and utilise GPUs should contact the team: https://neodaas.ac.uk/Contact