Tracking the Carbon Cost of Optimization Algorithms: A case study

One of the biggest global challenges is tackling climate change and building a sustainable future. The IT sector is estimated to account for 2-6% of global carbon emissions and is set to rise as much as 20% by 2030. Whilst software on its own doesn't emit CO2, understanding the cost of running it is important so that the way it is developed, and the way it is used, can be optimised for minimal carbon emissions. There are several carbon tracking tools available for estimating hardware electricity power consumption, allowing users and developers to quantify the carbon cost of running a program and identify areas for improvement. FitBenchmarking (https://fitbenchmarking.github.io/) is an open source Python package that allows users to compare the performance of different optimization algorithms and their implementations across various fitting problems, through metrics such as runtime and accuracy. In this talk, I will outline how a new carbon emissions metric has been added to FitBenchmarking, some of the challenges faced along the way, and the insight that reporting emissions offers developers and users of fitting software. A recording of this session is available on YouTube: https://youtu.be/lEkYZxXJ3Kk