Google Deepmind doubles down on AI-led efforts to improve datacentre energy efficiency

Deepmind is expanding its efforts to improve datacentre efficiency using artificial intelligence (AI) by giving these systems even more control over the cooling infrastructure in Google’s server farms.

The Google-owned artificial intelligence company has confirmed it is continuing to explore how using AI systems can achieve even greater datacentre efficiency gains, beyond what is possible by relying on human intervention alone.

As reported by Computer Weekly in 2016, this work has previously seen the organisation create a neural network-based system that is trained using data collected from sensors both within and outside of Google datacentres to track how various environmental factors impact on its performance, and predict the impact they may have on the site’s future energy consumption.

The system would then provide datacentre operators with a series of recommendations they could implement to improve the performance of its cooling system and reduce the amount of total power consumed.

When the company deployed the system at a live Google datacentre, a 40% drop in the amount of energy needed to cool the facility was reported, and contributed to the site achieving the lowest power usage effectiveness (PUE) score in its history of 1.06.

On the back of this success, Deepmind has taken the project one step further – the company revealed in a blog post – by putting the AI system to work directly on the cooling system, and scaling back even further the amount of human intervention needed.

“Instead of human-implemented recommendations, our AI system is directly controlling datacentre cooling, while remaining under the expert supervision of our datacentre operators,” the blog post stated.

The setup is already in use in “multiple Google datacentres”, the company said, and has been warmly welcomed by the search giant’s server farm operator community.

“The idea evolved out of feedback from our datacentre operators who had been using our AI recommendation system. They told us that although the system had taught them some new best practices … implementing the recommendations required too much operator effort and supervision,” the blog post stated.

“Naturally, they wanted to know if we could achieve similar energy savings without manual implementation.”

On this front, since the system was deployed a “matter of months” ago, Deepmind said it is already delivering consistent energy savings of around 30%, with further reductions expected in due course.

“That’s because these systems get better over time with more data,” the blog post continued. “Our optimisations boundaries will also be expanded as the technology matures, for even greater reductions.”

Looking ahead, the company said it is also looking at other use cases for the technology, beyond the walls of Google’s datacentres. “We’re excited that our direct AI control system is operating safely and dependably, while consistently delivering energy savings. However, datacentres are just the beginning,” the post continued.

“In the long term, we think there’s potential to apply this technology in other industrial settings, and help tackle climate change on an even grander scale.”