Time to stop using PUE to measure data centre efficiency?

Are the days of using PUE to measure data centre efficiency numbered? Is it time for a change?

Power Usage Efficiency

The data centre industry has long used the PUE as a quick and easy way to establish the energy efficiency of a given data centre.

A lower PUE is better; a theoretical PUE of 1.0 indicates perfect efficiency with no wasted energy within the data centre.

A flawed metric

Unfortunately, the PUE is riddled with limitations and doesn't provide a complete picture, especially for modern AI data centres.

Specifically, the PUE:

• Is often worse at lower IT workloads.
• Focuses only on facility efficiency.
• Can be manipulated.

PUE can be manipulated and abused in several ways.

• By excluding certain data centre subsystems or with selective reporting using only data from periods of high IT workloads.
• I've also heard of large cloud providers cramming their racks full of fans for cooling - but due to how PUE is calculated, means it gets counted as IT power.

PUE in legislation

On the other hand, governments in the region and beyond are using the PUE as a benchmark to determine data centres' sustainability.

For example:

• China plans to bring the average PUE of its data centres to below 1.5 by 2025.
• Singapore's IMDA hopes to lower PUE to less than or equal to 1.3 over the next 10 years.

Considering the ways it can be gamed at the expense of genuine sustainability, should we use other metrics instead?

What do you think?

Is measuring the PUE of data centres a shell game that's past its time? Is it time for a change - what should we use instead?

Are you aware of other ways PUE metrics can be gamed? Do share in the comments below.