Energy performance: look beyond a single PUE

Power Usage Effectiveness is the ratio of the total energy a data center consumes to the energy delivered to its IT equipment, measured over the same period and within a defined boundary. The Green Grid’s guidance sets out how the measurement is made and reported. It measures the overhead of the facility relative to its IT energy, and the sections below set out what it does not measure.

The arithmetic, and the two denominators

A PUE of 1.2 means that for every 1 kWh delivered to the IT equipment, the facility consumes 1.2 kWh in total. The overhead of 0.2 kWh is 20% of IT energy, or approximately 16.7% of total facility energy. Both are correct; they describe the same overhead against different denominators, and confusing them changes the meaning of a target by a sixth. When a tender, a lease or a lender’s covenant quotes a PUE, the first thing to establish is which arithmetic it is using and over which period.

Ask what sits behind the value

A quoted PUE is either a design estimate or a measured result, and the two should never be compared without saying so. For a design estimate, ask for the load at which it was calculated, the climate data used, the economiser hours assumed and the operating strategy of the plant: how many chillers run at part load, at what chilled-water temperature, with what fan and pump control. For a measured result, ask for the reporting period, the meter boundaries, the IT load during the period and the weather. A figure measured in a mild winter at full load is not comparable with one estimated at the design ambient at half load.

For a phased facility, ask for the expected performance at the first operating load as well as at full occupancy. Most facilities spend their first years at a fraction of their design load, where fixed losses dominate the overhead: transformer magnetisation, UPS no-load losses and the minimum turndown of the chillers. The equipment’s operating strategy at part load matters more to the number than its rated efficiency at full load, and it is the part of the design most often left to the controls contractor.

What PUE does not measure

PUE describes the facility’s overhead relative to the IT energy. It does not measure the useful computing work the IT delivers, so a hall full of idle servers can report an excellent PUE. It does not measure the carbon intensity of the electricity, so a facility on a coal-heavy grid and one on hydro can report the same figure. It does not measure water, so an evaporative cooling strategy that improves the PUE can consume millions of litres in a region where water is the scarcer resource. And it does not measure embodied impacts: the carbon in the concrete, the steel and the plant.

A design review therefore treats these as separate questions with separate answers:

• How will the power and cooling plant be staged as the load grows, and what does the overhead look like at each stage?
• What are the water requirements of the cooling strategy, and what are the local supply constraints and their price?
• Which metering arrangements will allow the operator to verify the performance, by hall, by system and by period?
• How are the energy objectives balanced against the resilience objective and the maintenance regime, which both consume margin?

Design so that the number can be verified

A target without a measurement plan is difficult to manage and easy to dispute. Agree the meter locations, the data ownership, the reporting periods and the operating assumptions while the design is being developed, so that the electrical power monitoring, the building management system and the DCIM are specified to measure what the target refers to. A PUE calculated from the utility bill and an estimate of the IT load cannot be verified by hall, by system or by period.

The aim is an energy strategy the client can evaluate under the conditions the facility will experience and at the loads it will carry, with the water and carbon consequences stated alongside it.