Google disclosed the improvement in a report by Bloomberg. The reduction is significant and has allowed Google to save substantial amounts of money. In 2014, it used over 4.4 MWh of power, the same amount as the average yearly consumption by 367,000 U.S. family homes. Reducing this figure by 40 percent has led to large savings.
The achievement has also allowed Google to pay for its acquisition of DeepMind in an unusual way. The technology it has acquired from the company has effectively paid for itself. By reducing the power bill, Google has generated enough money to pay for part of DeepMind.
The DeepMind AI now has control of Google’s datacentre cooling systems. Using neutral networks to make decisions, the AI turns on fans, opens windows and tunes the efficiency of the data centre to optimise electricity consumption and match it to the current load.
The AI uses a similar technique to the DeepMind software that taught itself how to play old Atari video games. It assigns itself a score based on the efficiency of the datacentre. As power consumption is optimised, it awards itself points. The aim is to get the highest possible score, indicating the most efficient consumption.
DeepMind now monitors around 120 variables in Google’s datacentres. The results have surprised both companies, leading to a substantial reduction in energy usage. The AI has also reduced Google’s impact on the environment, lowering its carbon footprint by minimising the energy it consumes.
With the project deemed a success, DeepMind is developing ways to further tune its AI and obtain even greater efficiency in the future. The AI is now able to work out where it lacks insights into the operation of the datacentre. It can report back to DeepMind which prompts Google to install additional sensors, enabling the AI to better understand how power is being used.
In the future, DeepMind could lend its AI to other similar buildings. It could be used in factories, hospitals, warehouses and other large facilities to cut power usage. On a larger scale, DeepMind could one day monitor power distribution grids and help to optimise efficiency across the entire network.
Datacentres are used by Google and other technology companies to store the vast amounts of data generated by users on a daily basis. The huge number of servers inside and the systems required to keep them cool demand very high amounts of electricity, making datacentres some of the largest energy consuming buildings in the world.
Companies including Apple, Google and Facebook have pledged to power their datacentres using 100% renewable energy, helping to negate their impact on the environment. Using innovations like DeepMind’s AI, Google can continue to increase the efficiency of its server farms, benefitting its finances and the world. The artificial intelligence will be deployed in all of its datacentres by the end of the year.