Topic guide

Performance metrics

Explore publications on PUE and other data centre performance indicators. Before using a metric for comparison or design decisions, check its measurement boundaries, calculation method and required inputs against the original reference.

Sources
212
Downloadable PDFs
60
Years
2007–2026

All sources

73–96 / 212

Technical guideFree to read

International Energy Agency 2025

IEA Energy and AI Observatory (data tool)

Link to the observatory from the tool's energy-context panel and take current national and regional data center demand figures from it rather than hard-coding values.

  • The related PDF reports are CC BY 4.0 and downloaded separately.
Technical guidePDF

Open Compute Project Foundation 2026

Introducing Power Compute Effectiveness (PCE), a Provisioned Capacity Allocation Metric for the New Data Center Era (October 2026)

Compute PCE automatically from the design model: utility/transformer capacity as P_facility, peak cooling electrical input at design ambient, UPS/transformer/busway losses and auxiliary panel loads, all on the declared redundancy basis.

  • Motor horsepower converts at HP x 0.746; kVA ratings convert to kW at the declared power factor.
Technical guideFree to read

Cundall 2024

Is power or water best for cooling?

When a user selects a cooling topology, show a three-way comparison of energy (PUE), water (WUE) and carbon (CUE using local grid intensity) for the chosen site.

  • Danish hyperscale project with Fjernvarme Fyn: heat pumps lift to 70-75 C, expected 165,000 MWh/yr of heat at full build-out.
Technical guideFree to read

Uptime Institute Journal (Uptime Institute) 2019

Is PUE actually going UP?

In a PUE benchmark display, show the survey average next to the hyperscale range.

  • Average PUE rose from 1.58 in 2018 to 1.67 in 2019 (sample of 624).
Research paperPDF

International Conference on Learning Representations (ICLR 2024) 2024

LLMCarbon: Modeling the End-to-End Carbon Footprint of Large Language Models

In AI data-centre carbon estimation, compute operational (FLOPs, power, PUE, grid intensity) and embodied (chip area × CPA, memory and storage per GB) carbon together; in a renewable-powered facility embodied carbon rises to 92-95%.

  • Against published Google and Meta figures, error is 8.2% or less for training and -3.05% for embodied carbon.
Research paperPDF

arXiv preprint (cs.SE); Skyworkz / Sustainable Digital Infrastructure Alliance (SDIA) 2023

Measuring IT Carbon Footprint: What is the Current Status Actually?

Compute IT carbon as (Es + Eo) × I + M and allocate embodied carbon (M) for shared resources by time and resource share; define scope, idle-time and shared-resource attribution explicitly.

  • Operational carbon = PUE × sum over intervals of grid-mix intensity × energy, constrained by energy-mix data granularity (for example 30-minute updates); average versus marginal emission factors are discussed.

All content and rights belong to the respective authors and publishers. SoSo CAD provides only the citation, its own summary and a link to the source, and is not responsible for the sources' content.

Monthly newsletter · free

Keep your reading list current

Once a month: new library sources, changes to Turkish regulations and upcoming data centre events.