Is PUE actually going UP?

Overview

The article starts from the finding of the 2019 Uptime Institute survey that average PUE rose from 1.58 in 2018 to 1.67 in 2019. The sample covered 624 sites. The author asks whether the rise reflects a real loss of efficiency or a measurement and sampling effect, and lists possible explanations. The article should be read as a short, open-ended assessment that questions how averages of a widely accepted industry metric should be interpreted.

The suggested causes include hotter weather, lower utilisation and higher power density. The piece stresses that these factors were not measured individually, so the exact cause of the rise is unknown; the text is therefore commentary rather than proof. In other words, the rise has not been separated by cause with data, so the results should be treated as a list of factors to watch rather than as a firm cause-and-effect relationship in design decisions.

The most important distinction in the article is between average PUE per site and overall PUE per kW of IT load. As work shifts to cloud and large colocation facilities, the site average can rise while the load-weighted overall PUE falls. The author also notes that PUE is not the only efficiency metric and cautions against reading it in isolation.

As context, hyperscale and large colocation operators claim annual or design PUE between 1.1 and 1.4. The largest efficiency gains are described as having come between 2007 and 2013 from hot/cold separation, higher temperatures and cooling controls, after which progress slowed. The declared low values should therefore be read as the results of large facilities with particular operating conditions, not as a target every sector average could reach.

Key points

1. Average PUE rose from 1.58 in 2018 to 1.67 in 2019 (sample of 624).
2. Hyperscale and big colo operators claim annual or design PUE between 1.1 and 1.4.
3. The largest efficiency gains came in 2007-2013 from hot/cold separation, higher temperatures and cooling controls.
4. Possible causes include hotter weather, lower utilisation and higher density.
5. Average PUE per site is not the same as overall PUE per kW of IT load.
6. Load-weighted overall PUE is falling as work moves to cloud and large colos.
7. PUE is not the only efficiency metric and should not be read alone.

Using it in design

• In a PUE benchmark display, show the survey average next to the hyperscale range.
• State clearly that a site-average PUE is not load-weighted.
• When setting a design PUE target, treat utilisation and seasonal temperature effects as separate scenarios.
• Present PUE together with complementary metrics in efficiency reports.
• Treat hot/cold separation, higher operating temperature and cooling controls as proven sources of PUE improvement.

Who should read it

Data center designers and energy managers who set efficiency targets or interpret PUE data.

Publication details

Authors

A. Lawrence

Publisher

Uptime Institute Journal (Uptime Institute)

Year

2019

Document language

English

Source type

Industry body

Access

Free to read

Licence

Copyright retained; citation only — the publisher’s terms apply to the original document

Citation