11. Outputs and Interpretation

11.1 Primary Outputs

The methodology produces the following metrics for each unit of cloud usage:

• Compute electricity, expressed in kWh. The estimated electricity drawn by the processor, its chassis, the host memory and drives attributable to the unit of cloud usage, and any attached accelerators.
• Compute electricity per usage unit, expressed in kWh per SKU usage unit. The same figure expressed against the unit the SKU is billed in, for application to a quantity on a cloud usage report.

11.2 Intended Use

These outputs are intended to support organisational sustainability reporting under frameworks such as the Corporate Sustainability Reporting Directive, Streamlined Energy and Carbon Reporting, and the climate disclosure rules of the US Securities and Exchange Commission; Scope 3 emissions measurement; instance family and processor selection; rightsizing and workload placement decisions; and comparative analysis of compute efficiency across instance families, processor generations, and cloud providers.

11.3 Interpretation Guidance

Outputs are estimates calculated from published server measurements, published hardware specifications, and billing data, and reflect the best available information at the time of calculation. Actual consumption in production cloud environments may vary.

Five points bear on how an individual figure is read, being the fixed operating point, the derivation used for a processor a cloud provider designs for its own fleet, the proportion of rated power taken for accelerators, the drive idle draw, and the modelled host configuration.

A figure is calculated against a fixed operating point. That point states nothing about the processor utilisation of the customer's own workload.

A figure for an instance built on a processor a cloud provider designs for its own fleet comes from the measured curve of a related processor, scaled by clock speed and by a published per-core factor. Its uncertainty comes from that factor and from the clock scaling. Greenpixie calculates this way every AWS Graviton generation, Azure Cobalt 100, and Google Axion in both the C4A and N4A machine series. The most recent generation of a cloud provider's own processor design has the fewest related processors with published measurements. Its figures carry the widest uncertainty of the processors calculated this way. Ampere Altra and NVIDIA Grace are not in that group, as the published measurements cover both and their figures come from those measurements like any other.

Accelerator electricity is calculated from a single proportion of rated power, measured on an inference workload. The figure for a training or graphics workload is therefore understated.

A drive is counted at its idle draw. A heavily used drive therefore draws more than the figure states.

The host configuration a figure is apportioned against is modelled, in both its memory modules and its drives.

Against a figure scaled from a related processor, Greenpixie records the published measurement it was built from and the processors it was scaled through. Against a figure taken from the median of comparable matched instance sizes, Greenpixie records how it was produced.

Compute electricity is one component of IT Electricity Usage. It is not a total for the workload. A workload that also uses block or object storage services carries storage this methodology does not count, as stated in Section 2.2.

11.4 Non-Intended Use

The outputs should not be treated as certified measurements, contractual guarantees, or regulatory compliance instruments unless explicitly agreed in writing. They should not be read as a measurement of an individual customer's processor utilisation, or as the storage footprint of a workload that also uses block or object storage services. They should not be used as the sole basis for carbon offsetting calculations without independent verification.