Using FPGAs to minimise data centre power consumption

The rising popularity of cloud computing, virtualisation and remote storage is fuelling increased investment in data centres. By 2018 global data centre traffic is expected to reach 8.6 zettabytes - the equivalent of 9 trillion hours of high definition video streaming – which is more than double the 2013 total of 3.1 zettabytes.

As data centre traffic rates have increased, interest in minimising the costs, both financial and environmental, has gathered momentum. It would take the annual output of 50 large coal-fired power stations to provide the projected 140bn kilowatt-hours of electricity required to power U.S. data centres in 2020 and the environmental impact of this has been compared with that of air travel. These costs not only present a challenge for a sustainable future, they also represent a significant financial burden on data centre operators; a burden compounded further by rising energy prices.

With the spotlight on these issues methods of reducing energy expenditure can be divided, broadly, into two categories: more efficient servers and more efficient systems for cooling them. This whitepaper focuses on the former method by looking at how a large proportion of data centre processing burden can be offloaded to hardware accelerator cards to reduce host CPU load and cut overall energy requirements.

Read and download the full whitepaper below.