Google: data center controlled via AI consumes 30 percent less energy on average

Google reports that a few months ago it switched to a system that uses artificial intelligence to manage the cooling of its data centers without human intervention. That would have resulted in an average energy saving of 30 percent so far.

DeepMind, a sister company of Google, writes in a blog post that the AI ​​system pulls in a snapshot of thousands of sensors from an entire data center every five minutes. It then analyzes how it can minimize data center energy consumption by taking certain actions delineated by security considerations. Judging by a graph, this ai control mode has been in use for about nine months. The energy savings of an average of 30 percent would have been achieved in that time. Further improvement would be possible due to more and more data.

The company does not disclose in which data centers control is left entirely to artificial intelligence, but speaks of “some Google data centers” where services such as YouTube, Gmail and Search are hosted. The system uses a degree of uncertainty, which means it will only take a certain action if it is sure it is the right one. In addition, additional checks are made to ensure that the promotion falls within the applicable safety standards. Finally, it is always possible to return control to traditional systems.

The idea of ​​leaving control entirely to AI came from the data center administrators themselves, who were already using the system to make recommendations. However, following those recommendations would require too much effort and oversight. DeepMind does not explain exactly what actions the system can take. Dan Fuenffinger, one of the Google data center employees, cites an example: “It was fantastic to watch the AI ​​learn to take advantage of the winter weather to produce colder water than usual, reducing the energy needed for cooling. .”