Modelling the Power Consumption of Scientific Computing

Computational work is a fundamental component of scientific research. However, whether run on a laptop or a supercomputer, this work requires the consumption of electrical energy. Generally, this is generated at least in part from fossil fuels, making computational work one of the major ways that scientific research contributes to climate change. In order to improve the sustainability of scientific computing, it is necessary to develop models to reliably measure and predict the power consumption of computational work.

To place these models on firm empirical grounding, it is important to collect granular data on how computing systems use power. To achieve this, we have built a computational test bench, effectively a desktop computer instrumented to enable monitoring of the power rails feeding individual components under various scientific workloads and benchmarks.

The models developed based on this data will be used to enhance the accuracy of publicly available tools, such as the Green Algorithms online calculator. Improvements to such tools can aid researchers in assessing the environmental impact of their computational work, encouraging more judicious use of computational resources, and promoting the development of more sustainable scientific computing practices.