“Also in real time we’re adjusting the parameters of the cooling system so that it more closely matches the demand from the compute processing”.
Advertisement
The results speak for themselves, with DeepMind’s program being able to keep cooling energy use 40% below normal levels.
The system cut power usage at data centres by several percentage points, which has an impact both on cost and the environment, so it’s a win-win situation.
Speaking at an AI conference in NY recently (via Bloomberg), Demis Hassabis explained that DeepMind’s technique to saving power in data centres was actually similar to the technique used by the software to teach itself to play Atari video games. A significant proportion of Google’s spending on electricity comes from its data centers, which support its globe-spanning web services and mobile apps. The company claims responsibility for 0.01 per cent of global electricity use. A whitepaper is reportedly to come from them within four to six weeks.
Fed vast swathes of historical data collected by sensors measuring temperature and power, DeepMind’s general goal algorithm – similar to the one trained to play Atari games – had come up with a real-time, adaptive system that had cut the cost of cooling by 40% and the overall energy consumption by 15%, Mr Suleyman said.
Inside a Google data centre. Now that DeepMind knows the approach works, it also knows where its AI system lacks information, so it may ask Google to put additional sensors into its data centers to let its software eke out even more efficiency.
“Conventionally a human manually tweaks a lot of the knobs that control the operation of the data centre”, Suleyman explained.
This is because the AI uses a system of neural networks to predict data centre behaviour, and boosting efficiency by not wasting energy on redundant processes. Each neural network was fed data on temperatures, power usage, pump speeds and more.
“Possible applications of this technology include improving power plant conversion efficiency (getting more energy from the same unit of input), reducing semiconductor manufacturing energy and water usage, or helping manufacturing facilities increase throughput”, they suggest.
Advertisement
The internet giant is using technology from the DeepMind artificial-intelligence subsidiary for big savings on the power consumed by its data centers, according to DeepMind co-founder Demis Hassabis. “The goal of these predictions is to simulate the recommended actions from the PUE model, to ensure that we do not go beyond any operating constraints”.
