by Johannes Schlippe, Senior Consultant at BusinessCode
At an all-day workshop in December, BusinessCode looked at how support from various AI applications can be skilfully integrated into the work of software developers.
I was interested in the environmental/climate aspects of this paradigm shift and therefore discussed estimates of energy consumption with ChatGPT. (I did not check the facts and sources for the following statements separately, as I found the statements essentially plausible).
How much energy does a software developer use?
In an eight-hour working day, a software developer probably consumes between 1.3 and 2.0 kilowatt hours of (electrical) energy for operating a laptop and peripheral devices as well as internet searches, assuming the use of ‘traditional’ (not explicitly AI-supported) search engines. (With a desktop computer instead of a laptop, consumption is almost twice as high).
How much energy do my AI queries consume?
I then asked how energy consumption increases when the person makes heavy use of requests to AI assistants (200 requests per day): namely around 44 kilowatt hours, about the same as a typical refrigerator consumes in 1.5 days.
With a moderate use of AI requests (50 requests per day), the energy consumtion is a quarter of the aforementioned figure, approximately 11 kilowatt hours (around eight times the ‘traditional’ consumption). This corresponds to a microwave oven running for around 10 hours (at maximum load). And assuming that the electricity consumed (on the part of the person and the AI data centre!) was generated from predominantly renewable sources, the additional carbon dioxide emissions are equivalent to burning one litre of petrol. With the current globally averaged electricity mix, which relies heavily on fossil fuels, it is 2.3 litres.
The generation of images by AI is particularly energy-intensive. The creation of a single image alone, such as the one for this blog post, can consume up to 0.011 kilowatt hours of energy – about the same as a full charge of a smartphone. If 1,000 images were created, the energy consumption would be equivalent to driving an average petrol car for around 6.6 kilometres (4.1 miles).
Final Thoughts
With these comments, I am by no means trying to denigrate the use of AI applications in our everyday work – but I would like to remind you that the great possibilities of these tools have a price (and I am not talking about licence costs) that we all ultimately pay. I can now better understand the magnitude of this ‘price’ with the above estimates.