The datacenter market is on the move – as is the corresponding power consumption. However, conventional metrics for energy efficiency do less and less justice to reality because they do not consider the biggest power guzzler in datacenters: IT. A preview of the keynote speech of our president, Babak Falsafi, at Data Centre World in London.
As the backbone of the digital world, datacenters are central to discussions on energy use and environmental sustainability. Growing global scrutiny has led to regulations or moratoriums on new server farms in countries like Ireland, Germany, Singapore, and China, driven by concerns over their energy consumption.
Schneider Electric predicts a 5% annual rise in IT sector electricity consumption from 2023 to 2030, with 75% driven by datacenters (due to AI) and mobile networks (transitioning to 5G). Meanwhile, a 2024 Uptime Institute survey reveals only slight improvements in datacenter energy efficiency, as measured by the conventional Power Usage Effectiveness (PUE) figure, which dropped from 1.65 in 2013 to 1.56 in 2024.
What are the limits of PUE?
Like many other key figures, PUE has its limitations: It neglects various ways in which the overall energy flow in the datacenter can reduce emissions, such as waste heat recovery and renewable energy. And it is subject to fluctuations depending on factors such as the time of year, the current workload of the datacenter and even the time of day – making it an unreliable measure of efficiency.
Its biggest flaw, however, is its inability to measure IT efficiency. Ironically, inefficient servers can lower PUE values because higher IT energy consumption improves the metric, incentivizing surplus IT resource allocation. The utilization of IT systems also has a significant impact on operational efficiency – but PUE does not indicate whether servers are being utilized at 20% or 80% of their capacity. The question of electricity consumption in datacenters and its efficiency therefore revolves around IT.
What are the right metrics for IT?
With rising IT power consumption, new metrics are needed to evaluate datacenter energy efficiency and emissions, covering both infrastructure and IT equipment. These should account for heat recycling, renewable energy use, and the efficiency of IT components such as CPUs, GPUs, accelerators, memory, storage, and networking equipment.
Workload utilization, for example, provides a more nuanced insight into computer resource efficiency. A server utilized at just 20% most of the time represents wasted resources, leading to unnecessary energy consumption and increased operating costs.
Technological excellence can also serve as a benchmark for datacenter efficiency. For instance, opting for flash storage instead of hard disk drives significantly lowers power and cooling needs while boosting data access speed. Similarly, fiber-optic cables improve network speed and reduce energy consumption compared to copper.
And finally, the maximum operating temperature is another key efficiency indicator. While datacenters were traditionally kept at low temperatures to avoid overheating, modern devices can operate safely at higher temperatures. Raising these limits reduces cooling energy needs, significantly improving overall efficiency.