What is power usage effectiveness (PUE) and data center energy efficiency?

Power Usage Effectiveness (PUE) is a well-established way to quantify the energy efficiency of a data centre – but does it tell the full story with server racks and IT equipment becoming increasingly efficient? Is there a better measure and does PUE limit the ICT industry’s ability to support broader goals like reducing carbon emissions? How does it address changes in technology such as direct-to-chip liquid cooling?

In a world where data centres are an increasingly critical component of global digital enablement, we must explore whether there is a more complete measure of energy efficiency that supports the industry’s ability to reduce carbon emissions.

Key takeaways

• PUE as a standalone measurement presents limitations in measuring the data centre efficiency.
• New technologies such as artificial intelligence (AI) and liquid cooling challenge the effectiveness of PUE as a standalone metric.
• There is a growing need for other measures that account for both infrastructure efficiencies and IT productivity to support broader goals like reducing carbon emissions, as well as to ensure the data centre infrastructure is correctly sized.

How PUE is measured and why it’s flatlining

PUE is a simple metric that is used to compare data centre efficiency and is calculated by dividing the data centre’s total annual energy consumption by the total energy consumed by the IT equipment. The closer the figure is to a perfect ‘1’ the more energy efficient the data centre.

The global average PUE for data centres has largely flatlined in recent years, averaging between 1.55 and 1.59, reflecting an industry that has reached diminishing returns on efficiency improvements like optimising cooling systems and power distribution. But with large increases in AI loads and liquid cooling seeing a massive resurgence, a PUE measurement is not accurate on its own. Paradoxically, with more efficient server racks and the resurgence of liquid cooling, the PUE rate shows an increased figure rather than a reduced result.

Primary goal of power usage effectiveness

The chief aim of (PUE) is to alleviate operational costs and enhance environmental sustainability by reducing energy consumption in data centers. This goal is more relevant than ever in a world increasingly focused on green initiatives. Data centers are vital players in the digital economy, but their energy usage can contribute significantly to environmental degradation.

Reducing PUE can lead to more economical operation of data centers, offering cost savings and providing a competitive advantage by improving operational efficiency. It’s worth noting that a lower PUE number denotes a more energy-efficient data center. In an ideal world, the lowest PUE is 1.0, indicating that the servers use all the energy going into the data center. However, energy is also needed for lights, cooling, and other necessary functions, making achieving a PUE of 1.0 difficult, if not impossible.

Components of power usage effectiveness

Data center providers calculate PUE using IT equipment power, cooling infrastructure power, and the power used by lighting and other miscellaneous systems. Each component contributes to the overall PUE figure and provides a clearer picture of where energy is used within a data center, reflecting the total facility power.

• IT load: This is the actual computational equipment, including servers, storage arrays, and networking gear, that perform the data processing tasks. Efficient use of this equipment is crucial since it is the purposeful use of energy within a data center.
• Power distribution: The efficiency of power distribution systems, including transformers, uninterruptible power supplies (UPS), and power distribution units (PDUs), dramatically affects PUE. Any energy loss in power conversion or distribution adds to the overhead.
• Cooling infrastructure: One of the largest non-IT energy consumers is the cooling system. This includes chillers, cooling towers, economizers, CRAHs, and CRACs.
• Airflow management: Proper management of hot aisle and cold aisle containment and the use of blanking panels and floor layout can significantly improve cooling efficiency and reduce PUE.
• Building management systems (BMS): These systems monitor and control the environmental conditions. Smart BMS can dynamically adjust the cooling based on the IT load and external weather conditions to optimize energy usage.
• Lighting: Even though it usually accounts for a small fraction of the energy used, efficient lighting systems such as LED fixtures with motion sensors can contribute to a lower PUE.
• Support systems include the energy consumed by backup generators, fire suppression systems, and security systems. While these don’t directly contribute to IT operations, they are essential for data center reliability and safety.

Upon examining these components, data center operators can pinpoint high energy usage areas and initiate steps towards their reduction. For instance, they might upgrade to more energy-efficient hardware, implement more effective cooling systems, or improve their power distribution setup. In this way, understanding the components of PUE can inform strategies to improve data center efficiency and reduce energy consumption.

Rethinking data centre efficiency metrics

PUE has long been the gold standard and industry benchmark for measuring data centre efficiency, however it provides limited incentives for organisations to make their IT infrastructure and computing more efficient (since more efficient computing results in a worsening PUE score). As a result, when tenants explore options for data centres to house their servers, consideration is sometimes only given to how power efficient the data centre infrastructure is, rather than the power efficiency of the compute infrastructure.

The primary purpose of a data centre is to process data efficiently but as AI workloads and more sophisticated server technologies emerge, the impact is felt on both the numerator and denominator of the PUE equation, resulting in misleading efficiency readings.

While PUE is not entirely obsolete, it is insufficient on its own. In a data centre, roughly 80 per cent of energy runs computers, and roughly 20 per cent runs the supporting infrastructure. Focusing only on minimising the 20 per cent of energy consumption attributed to supporting infrastructure overlooks the potential savings from optimising the 80 per cent used by IT equipment.

Improving IT efficiency can significantly enhance overall data centre performance, reducing energy used, and this highlights the need for a broader set of metrics that address both elements of the energy equation. If you just improve the IT equipment efficiency in an existing data centre, you actually increase the PUE.