Google has been using artificial intelligence for a wide range of tasks, ranging from delivering search results to speech recognition, so perhaps it should come as no surprise that Google’s latest AI product was figuring out how to improve the energy efficiency of the very servers used to do all that other stuff.
A Google blog entry spotted by Engadget describes how a Google engineer used his 20 percent time to apply machine learning to predict the real-time energy efficiency of its data centers. Google uses a measure known as Power Usage Effectiveness (PUE): a ratio of total power used to power actually used for computing. In simple terms, if cooling used as much power as computing, the PUE would be 2. The closer to 1 Google can get, the more efficient the energy usage.
Google has already got its PUE down to 1.12 – about twice as efficient as a typical data center – but is using the AI project to try to further reduce the number. By using machine learning to predict the impact of variables like outside air temperature, Google can tweak the setup to minimize power usage.
The days of self-aware machines grow ever closer …