Green Algorithms: Quantifying the Carbon Impact of Computations- Oxbridge Varsity Sci Symposium

Talk summary by Gemma Penson

Climate change is one of the most critical emergencies humankind and our planet face on a daily basis. Amidst the current pandemic, the situation is not getting any better, with  a severe reduction in recycling and mass production of disposable masks. For anyone determined to reduce their environmental damage and who is passionate about science, one of the computing talks given by Oxbridge’s Varsity Scientific Symposium offers an interesting perspective.

Hosted by Loïc Lannelongue, a Health Data Scientist at Cambridge University, the talk explored quantifying and raising awareness about the amount of carbon dioxide emitted by algorithmic computations. The project, which aims to make algorithms ‘greener’, was born after the shock caused by the Australian bushfires in January 2020 and has resulted in an open source, online calculator[1] being created.

“…the impact of computations had been widely unappreciated, if not unthought of by the public.”

Performance metrics of research projects are rarely undisclosed— so why aren’t the carbon footprints of these projects reported, too? For many years, the consensus in the environmental science community was  that it wouldn’t be possible to estimate this carbon footprint and the impact of computations had been widely unappreciated, if not unthought of by the public. A staggering 100 megatons of CO2 emissions are contributed by data centres and computing facilities every year, similar to those produced by American commercial aviation.[2] But it isn’t just large companies contributing to this problem, as each Google search you perform also releases roughly 10g of CO2[3] into the atmosphere. With 70,000[4] Google searches being performed every single second, the digital age is a significant contributor to global warming.

In the talk, Loïc discussed his mathematical model, which can be used to  calculate the amount of energy a computation uses in order to define its environmental impact. The amount of energy required is dependent upon a number of hardware specifications, such as  the runtime and the type and number of cores used.

A very insightful fact is that the size of the memory available, rather than the amount of memory used, is what affects the CO2 emissions of the computation. Thus, by ‘playing it safe’ and requesting more memory than you think you’ll need, you are unnecessarily increasing your carbon footprint with no additional computational gain.

Following Loïc’s mathematical model, once the required amount of energy  has been calculated,  itis multiplied by a carbon intensity factor to yield the environmental impact of the computation. This factor takes into account the global location of the computation  and the technologies used to produce the energy required.

Location is one of the most severe and dangerous factors in computational impacts, as running the same software in Australia will produce an 81-fold increase in CO2 emissions compared to when it is run in Norway! This tremendous difference is a result of Australia predominantly using coal as their main power source, whereas Norway relies on hydroelectricity produced from the rivers that cascade into its fjords[5]

Once the carbon footprint of a computation has been determined, the online calculator displays emission comparisons with car and plane travel. Loïc and his team also created their own comparison metric, called a ‘tree month’, which signifies how long it would take for a single tree to absorb the emissions of the respective computation. This unit easily allows businesses to counteract their carbon footprint by planting trees which will  absorb about 1.14kg of CO2 each, every month, a quantity equivalent to a single tree month.

Image via The coloured striped background, created by Ed Hawkins[*], represents the change in world temperature from 1850 to 2018.

[*] Hawkins, Ed. Institute for Environmental Analytics. #ShowYourStripes. [ONLINE] (2020).

So how can you help?

As location is one of the most significant factors, you should opt for a data centre in a country with low carbon intensity such as Sweden, Switzerland or France. Although it’s not always the case, many cloud providers will offer this option.

In addition, only request the amount of data you will use, aim to produce optimal algorithms and, above all, only run jobs that are actually required!

If you are willing to make your impact go further, getting involved in tree plantation is an excellent way to counteract COemissions.

A movement I strongly support is Team Trees— a fundraiser started by YouTuber Mr Beast and supported by the digital community. #teamtrees raised funds to plant an incredible 20 million trees in under two months and each $1 donation today enables another tree to be planted[6].

This talk offered an insightful perspective about the extent of our digital environmental impact iand the rarely-thought-of consequences of computational calculations. Loïc’s mathematical model is a very useful tool, encouraging Computer Science students like me to report the full CO2 emissions of their projects alongside other common metrics.

[1] Lannelongue, L., Grealey, J. & Inouye, M., Green Algorithms: Quantifying the carbon emissions of computation. arXiv:2007.07610 (2020)

[2] Jones, N. How to stop data centres from gobbling up the world’s electricity. Nature 561, 163– 166 (2018)

[3] Youtube. Varsity Sci – Day 3 – Scientific computing. (2020)

[4] Google Search Statistics – Internet Live Stats (2020)

[5] Carroll, Matt. National Geographic. Norway leads the charge on a sustainable electric future. [ONLINE] (2019)


This article reports on the talk hosted by Loïc Lannelongue, Health Data Scientist at Cambridge University on the 23rd of September- Day 3 at the Varsity Sci Symposium 2020.