# Eligibility, Selection, and Inclusion Logic | Embodied Emissions methodology
Source: https://greenpixie.com/docs/methodology/cloud-embodied-emissions/eligibility-and-inclusion

What is included and how it is selected and mapped.

## 6.1 Usage Eligibility [#61-usage-eligibility]

Greenpixie calculates embodied emissions for each unit of cloud usage that occupies hardware capacity. Support charges, reserved capacity fees, and other usage that occupies no hardware capacity are retained and categorised. Greenpixie calculates no embodied emissions for them.

## 6.2 Instance Resolution [#62-instance-resolution]

Greenpixie first matches each billing usage type to the instance it bills for. Only then is the hardware of that instance identified. A cloud provider's billing catalogue lists the same instance under several usage types, covering on-demand, spot, dedicated, and the managed services that run the same hardware. Greenpixie matches every one of those usage types to the same underlying instance.

## 6.3 Component Identification [#63-component-identification]

Each resolved instance is mapped to the components of the host that serves it. Those are the processor model and the dies in its package, the memory capacity and the number of modules that capacity is spread across, the drives attached to the host, and any accelerators.

A cloud provider does not publish the configuration of the physical host an instance runs on. Greenpixie therefore models the host as running instances of the size the figure is being calculated for, as many as the processor supports, with the host's memory slots populated. Section 11.3 states how a figure derived this way should be read.

## 6.4 Component Generation Selection [#64-component-generation-selection]

Memory and drive emissions both depend on the generation of the die inside the part. Where the memory generation of an instance is known from the cloud provider's own documentation or from the processor specification, the embodied emissions for that generation are calculated. Where it is not known, an average is taken across multiple memory generations.

For solid state drives, the NAND flash generation is inferred from the launch year of the instance's processor, on the basis that the processor and the drives of a server are procured together.

## 6.5 Accelerator Memory Selection [#65-accelerator-memory-selection]

Several accelerators are sold as multiple variants with different specifications, including different amounts of high bandwidth memory. Those variants have different embodied emissions figures. Greenpixie maps each accelerator-attached instance family to the memory capacity the cloud provider sells with it, taken from that cloud provider's own family, machine family, or size series documentation.