Compute

2. Scope and Applicability

What is in scope, out of scope, and the evaluation boundary.

2.1 In Scope

The methodology applies to the following:

  • Supported cloud providers. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP).
  • Compute hardware components. The electricity drawn by the processor, the server chassis that houses it, the memory modules populated on the host, the drives attached to the host, and any accelerators attached to the unit of cloud usage.
  • Local storage. The drives sold with a compute instance and physically attached to its host, being the instance store volumes on AWS, the temporary and local NVMe disks on Azure, and Local SSD on Google Cloud. Each is an NVMe solid state drive, a SATA solid state drive, or a mechanical hard disk drive.
  • Instance-based compute. Virtual machine instances from the three cloud providers, including burstable and shared-core sizes, accelerator-attached sizes, and sizes built on the cloud providers' own processor designs.
  • Serverless compute. Function and container services billed per vCPU-second or per vCPU-hour. These take a per-vCPU rate, as described in Section 7.6.
  • Billing unit coverage. Every compute usage type that resolves to an instance, including on-demand, spot, dedicated, reserved, and the managed service variants that run on the same hardware.
  • Reporting granularity. Electricity values are calculated for each unit of cloud usage, expressed in the unit that usage is billed in.

2.2 Out of Scope

The following are not covered by this methodology:

  • Block storage, object storage, and archival storage sold as their own services, being Amazon EBS and Amazon S3, Azure managed disks and Blob Storage, and Google Cloud Persistent Disk, Hyperdisk, and Cloud Storage. These carry their own usage types on a cloud usage report. Greenpixie produces an electricity figure for the storage sold with an instance. This methodology substitutes no estimate for the separately billed services.
  • Drive controller, backplane, expander, and drive shelf electricity, which is not modelled. Only the drives themselves are counted. The electricity figure for local storage is therefore a lower bound for the drives and the hardware that connects them.
  • Network device electricity. The parent cloud methodology defines IT Electricity Usage across four component categories, being computation, memory, storage, and networking. This document covers the computation, memory, and storage components of it for an instance.
  • Everything else the parent cloud methodology covers, being facility overhead, grid carbon intensity, water consumption, and the embodied emissions of the same hardware.
  • On-premises and private cloud infrastructure.
  • End-user device energy consumption.

2.3 Evaluation Boundary

The evaluation boundary encompasses server-level electricity attributable to the processor and its chassis, to the memory modules the host is configured with, to the drives attached to the host, and to attached accelerators, measured at the point of alternating current supply to the server and apportioned to the unit of cloud usage. Chassis overhead, being fans, voltage regulation, and power supply losses, is counted inside the electricity attributable to the processor.

On this page