Examples
Worked examples
- Is an instance
A 10-hour genome-assembly job on shared HPC estimated at roughly 5.6 kg CO2e using energy draw x grid intensity x PUE.
Counter-examples
Looks similar, but isn't
- Not an instance
Treating a lab's compute footprint alone as its total carbon footprint, when travel and procurement are usually larger contributors — see research carbon footprint for the full roll-up.
Editorial commentary
Compute footprint is calculated as (energy used in kWh) x (regional grid carbon intensity in kg CO2e per kWh), with adjustments for power usage effectiveness (PUE) of the data centre and embodied emissions of the hardware. Tools such as Green Algorithms, CodeCarbon, and the ML CO2 Impact calculator estimate footprint at job or experiment level. Large machine-learning training runs have become emblematic, with single transformer pre-training campaigns reported at hundreds of tonnes CO2e. Researchers are encouraged to choose low-carbon regions, schedule for low-carbon hours, and right-size compute jobs.
How this differs from the more specific compute pages
This page covers the general calculation method for any computing workload — a bioinformatics pipeline on a shared HPC cluster, a climate simulation, an image-processing job — not only AI. Two pages specialise it for machine learning: training carbon footprint covers the one-off (or periodically repeated) cost of training a model, and inference carbon footprint covers the per-request cost of serving a trained model — both need different formulas (per-run vs. per-token/per-request) than the generic kWh × grid-intensity approach here. Research carbon footprint is the higher-level roll-up: an institution- or project-wide total in which compute is one contributing category alongside travel, procurement, and facility energy. Sustainable HPC covers the facility-level operational and procurement decisions (cooling design, hardware refresh cycles) that determine the grid-intensity and PUE inputs this formula depends on, and energy proportionality is the underlying hardware/software design property that determines how efficiently a given job converts power draw into useful work in the first place.
Worked example
Running a 10-hour genome-assembly job on a shared HPC cluster drawing an average 2 kW, in a region with a grid carbon intensity of roughly 0.2 kg CO2e/kWh and a data-centre PUE of 1.4, gives an estimated footprint of approximately 2 kW × 10 h × 1.4 × 0.2 kg CO2e/kWh ≈ 5.6 kg CO2e for that run — the same arithmetic that Green Algorithms and CodeCarbon automate, regardless of what the job actually computes.
Counter-example
Reporting only a lab’s compute footprint and presenting it as “the lab’s carbon footprint” understates the total — for most research groups, travel and procurement outweigh compute; see research carbon footprint for the full-scope roll-up.
References
- Lannelongue, Grealey, and Inouye, “Green Algorithms: Quantifying the Carbon Footprint of Computation,” Advanced Science 8(12), 2021.
- CodeCarbon project documentation (codecarbon.io).
- ML CO2 Impact calculator (mlco2.github.io/impact).
Also known as
Compute GHG footprint · ML carbon footprint · HPC carbon footprint
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Compute carbon footprint"
vocab-term-identifier="https://casrai.org/dictionary/term/compute-carbon-footprint" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/compute-carbon-footprint",
"name": "Compute carbon footprint",
"identifier": "https://casrai.org/dictionary/term/compute-carbon-footprint",
"description": "The greenhouse gas emissions attributable to the electricity consumed by computational research activities, including local workstations, on-premises HPC clusters, and cloud or commercial compute services.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/sustainable-research#set",
"url": "https://casrai.org/dictionary/term/compute-carbon-footprint",
"sameAs": [
"Compute GHG footprint",
"ML carbon footprint",
"HPC carbon footprint"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:23:01",
"dateModified": "2026-08-23T05:19:53",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
