Data centre energy consumption

How to meet the artificial intelligence energy challenge

The growing energy footprint associated to data centres needs to become more efficient. A new JRC report presents three approaches.
Published: Fri 14 Dec 2018

In a recent article we discussed the double edged challenge of the growing use of artificial intelligence (AI), and how on the one hand it is leading to rapid growth in the need for energy hungry data centres but on the other how the technology itself is being harnessed to optimise their energy consumption.

A new policy study based on research and analysis by the European Commission’s Joint Research Centre (JRC) investigates this issue further, pointing to energy as “a main determinant” of the long-term sustainability of AI and urging “early action” to address the energy footprint in its development.

In doing so, two factors come into play. One is how data and its use is expected to grow in the future. According to the report there seems to be consensus that data growth will largely outpace foreseeable improvements in computational power and the associated energy footprint, with the current early development of 5G and the Internet of Things (IoT) only exacerbating this.

The second is the energy usage associated to data centres. In Europe the number of data centres has dramatically increased in the last 10 years, driven largely by the development of cloud computing, according to the JRC, which estimates that their consumption combined with that of the data transmission could account for 3-4% of all EU power consumption.

However, the organisation also notes that currently at least and notwithstanding the increase of data and processing, data centre consolidation, outsourcing and cloud computing are helping to keep their energy consumption flat. The larger centres tend to be more efficiently designed and managed.

Energy efficiency options

Stating that both the computing and energy consumption generated by AI and the IoT – including the devices, networks and data centres to which they are connected – must be optimised significantly, the report proposes three approaches to address these issues.

The first, and of most relevance to the sector, is to improve the energy consumption of the data centres. Options include replacing older servers with newer more energy efficient units and improving the efficiency of cooling by not over-cooling or exploiting outside air or water sources.

The second, which is being introduced increasingly in the smart grid, is to move the intelligence to the edge and thereby reduce the data transmission and central storage and computing requirements.

The third, outside the scope of the sector, is to improve computer processing capabilities with advanced technologies and architectures to process big data more efficiently. In addition, the energy embodied in the production of goods should be reduced as should the obsolescence of digital technologies, notably mobile phones, which the report considers “perhaps the main energy consumption voice for digital technology”.

Concluding on the need to accelerate the speed of energy consumption reduction, the report notes: “For that we need to carefully estimate the amount of energy required by AI (and the IoT), taking into consideration all the required phases – e.g. data generation, moving, storing, pro­cessing, and insight generation and provision."