Artificial Intelligence (AI) promises huge improvements in cloud-driven corporate data processing and decision-making. Google’s DeepMind AI unit has also reported major success in improving data center efficiency. The sooner these innovations can be shared and implemented more widely the better.

Eric Lisica, Operations Director, EvoSwitch

AI – and specifically deep learning based on large scale cloud-based data analytics – is being applied across global business with startling results. According to McKinsey, location-based services and retail have experienced the greatest impact so far, with location-based services extracting 60% of the value to date and US retail adopters who use AI making up to 19% more margin. However, take-up has been slower than anticipated by the analysts. Many organisations have access to huge amounts of data, but the task of structuring it for easy access and manipulation can be daunting.  Also, new AI generally requires a huge amount of analytical talent, which tends to restrict breakthroughs to more recently-established or better-funded firms.

IQ in the DC

One of these firms is Google. As well as developing algorithms which are great at playing games and can translate and recognise images more quickly, Google’s DeepMind machine learning programme can also reportedly improve the energy efficiency of data centers. Google’s  Data Center IQ (DCIQ) project involved building a proof-of-concept ‘neural-network ensemble’ which simulated the components in a data center. Once this was built, the team added DQN (‘Deep Q-Network’), which is a deep learning AI algorithm developed as part of DeepMind. When this new application was plugged into a data center, it managed a remarkable 40% reduction in energy used for cooling, which equates to a 15% reduction in overall Power Usage Effectiveness (PUE) overhead.

The DCIQ team’s work could also potentially improve power plant conversion efficiency (getting more energy from the same unit of input), and reduction in semiconductor manufacturing energy and water usage, or helping manufacturing facilities increase throughput. These sorts of improvements across the whole ICT manufacturing and usage chain are exactly the sorts of joined-up initiatives we have been pushing for in our recent sustainability white papers.

Shared Learning

DCIQ clearly has huge potential for other data center operators. All the more so as Google has stated its intention to share the technology. DCIQ team leader Jim Gao said, “We’re trying to be really open source about this. We strongly believe that the work we’re doing can benefit others as well.”

 

A new white paper on the subject will be released soon and we look forward to studying it and becoming part of the next phase of the machine learning revolution.

Further Reading

The Age of Analytics: Free report summary from McKinsey

More about Google DCIQ: https://environment.google/projects/machine-learning/

EvoSwitch White Papers: Data Center Sustainability: The Next Dimension, and Reuse & Recycling of Servers: Next Steps Towards Sustainability