Reducing the carbon footprint of data centres through energy-efficient AI

Researchers are replacing traditional processing units with field-programmable gate arrays to help improve the sustainability of data centres

Spotted: The use of artificial intelligence (AI) and machine learning is growing by leaps and bounds, but so is the energy it uses. In fact, there are increasing concerns about the sustainability of AI and big data in general, with MIT reporting that the cloud has a larger carbon footprint than the entire airline industry and that some data centres consume the same electricity as 50,000 homes.

However, AI could also help the process of decarbonisation. Making data processing more efficient is therefore an important goal, and a recent international research project headed by Professor Marco Platzner of Paderborn University has made an important breakthrough towards this end.

The researchers are attempting to significantly reduce the required number and accuracy of calculations used in statistical machine learning processes, a key component of AI. They do this through the use of field programmable gate arrays (FPGAs) in place of the usual central processing units (CPUs) and graphics processing units (GPUs).

FPGAs are reprogrammable hardware devices that optimise computing, memory, and interconnection networks. Using FPGAs allows a reduction in the number of individual calculation steps, without affecting the quality of the output. This enables the researchers to significantly speed up computing time, saving energy.

The research is being undertaken at Platzner’s Computer Science Department and the Paderborn Centre for Parallel Computing, which has one of the most powerful FPGA systems in the world. It is supported by the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection, which has awarded the researchers around €1.5 million in flagship funding.

Because computing is growing at such a rapid pace, it has become imperative to make data centres more sustainable. Other ideas for decarbonising data that Springwise has spotted in the archive include using more efficient power distribution units and powering data centres with hydrogen.

Written By: Lisa Magloff