Power Modelling

Understanding the power usage of scientific computing

The tools developed by our lab, such as the Green Algorithms online calculator, rely on predictive models to estimate the power consumption of computational work. To improve the accuracy, reliability, and validity of these models, we collect and analyse power usage data from a variety of sources.

One important source of data is from experiments run on an in-house computational test bench, purpose built for the collection of fine-grained power measurements under controlled conditions. We run a combination of synthetic benchmarks (A.K.A. stress tests) and realistic scientific workloads on the test bench while monitoring the power draw of individual compute components and the system as a whole. Results from these experiments are compared with data sourced from high-performance computing (HPC) clusters and publicly-available databases to ensure the accuracy and widespread applicability of the power models we develop.

Get involved in this work

Are you interested in understanding the energy usage of your scientific workloads? If so, we would be happy to run a provided workload on the test bench! You will receive a detailed report on the power consumption behaviour of your workload while simultaneously assisting us by providing data relevant across different scientific computing fields.

We are currently finalising a standardised protocol to enable a smooth benchmarking procedure and ensure consistency across submissions. While the full documentation is being prepared, please feel free to get in contact directly with Jack to discuss running your workload on the test bench!