Google cuts data-centre energy consumption by 15% with AI

Google has announced it has managed to cut data-centre energy consumption by 15 per cent by using artificial intelligence to determine where it needed to cut energy consumption.

The technology, which uses machine learning to work out how much energy was required for cooling and airflow regulation was developed by DeepMind, a company that Google purchased in 2014. The technology constantly adjusts air temperature, pressure and humidity to make the environment as energy efficient as possible.

According to DeepMind although humans were able to work these calculations out, using knowledge learned from real-world scenarios made the process much faster.

IT Pro quoted Mustafa Suleyman, DeepMind's co-founder, ''It's one of those perfect examples of a setting where humans have a really good intuition they've developed over time but the machine learning algorithm has so much more data that describes real-world conditions."

''It's much more than any human has ever been able to experience, and it's able to learn from all sorts of niche little edge cases seen in the data that a human wouldn't be able to identify. So it's able to tune the settings much more subtly and much more accurately.''

The AI had succeeded in shaving several percentage points off of data -centre energy consumption, Bloomberg reported. This had led to 15 per cent improvement in power-usage efficiency, the metric of how much power went into the actual computing as against auxiliary services at the data centers.

"[The AI] controls about 120 variables in the data centers. The fans and the cooling systems and so on, and windows and other things," DeepMind's co-founder, Demis Hassabis, told Bloomberg.

A few percentage points might not sound like much, but that was multiplied over a vast amount of energy. Google's 2014 energy consumption stood at 4,402,836 megawatt-hours, the overall company, not just the data centres. US data centres that year consumed 70 million megawatt-hours, which totaled 2 per cent of the nation's energy consumption. If DeepMind's (currently proprietary) technique became an industry standard, it could save millions of megawatt-hours each year in the US alone.