3. Definitions

Accelerator. A specialised processor attached to a compute instance for parallel workloads, most commonly a graphics processing unit.

Active Idle. The power a server draws while powered and running its workload software with no transactions to process, forming the lower bound of a measured power curve.

Capacity Point. A drive capacity for which a power figure is held, from which figures for intermediate capacities are interpolated.

Chassis Overhead. The electricity drawn by a server's fans, voltage regulation, and power supply losses, as distinct from the electricity drawn by the processor itself.

Chip Vendor. The company that designs and sells a processor, an accelerator, a memory part, or a NAND flash part. Distinct from the cloud provider that buys the part and installs it in a server.

Cloud Provider. A company that offers cloud computing services and solutions to businesses and individuals. In this methodology, the term refers to AWS, Azure, and GCP.

Compute Electricity. The estimated electricity drawn by the processor, its chassis, the host memory and drives attributable to a unit of cloud usage, and any attached accelerators, expressed in kWh.

Compute Electricity per Usage Unit. Compute Electricity expressed against the unit a SKU is billed in, in kWh per SKU usage unit.

Drive. A physical storage device installed in a server, being a solid state drive or a mechanical hard disk drive.

Form Factor. The physical package an accelerator is built in. Some chip vendors publish a separate rated power for each form factor of the same accelerator model.

Hardware Thread. The smallest unit of parallel execution a physical processor core presents to an operating system. A core with simultaneous multithreading, being a core that runs more than one thread at once, presents more than one.

Idle Draw. The electricity a hardware component, such as a processor, a memory module, a drive, or an accelerator, consumes while powered and available but not servicing work.

IT Electricity Usage. The sum of electricity consumed by the four component categories within the IT equipment of a data centre, being computation, memory, storage, and networking. Abbreviated as ITEU.

Load Level. A workload level on a measured power curve, expressed as a share of the throughput the server reached at full load. A load level does not measure processor utilisation.

NVMe. Non-Volatile Memory Express. An interface for solid state drives connected over the PCI Express bus.

Operating Point. The point on a power curve at which the electricity for a unit of cloud usage is calculated.

Per-Core Factor. The published change in the performance one core of a processor design delivers against one core of the design preceding it, at the same clock speed, as stated by the company that designs the core.

Power Curve. The relationship between a server's workload level and the electricity it draws, expressed as a series of measured points from active idle to full load.

Process Node. The semiconductor manufacturing generation a die is fabricated on, expressed in nanometres.

Rated Power. The thermal design power a chip vendor publishes for a processor or accelerator, being a design target for cooling. A rated power figure does not measure draw.

SATA. Serial Advanced Technology Attachment. An interface used by mechanical drives and by lower-throughput solid state drives.

Server Vendor. The company that builds and sells a complete server and submits it for power measurement. Distinct from the chip vendor whose processor it installs, and from the cloud provider that operates it.

SKU. Stock Keeping Unit. In cloud billing, a distinct combination of service, resource type, region, and pricing tier that appears as a line item on a cloud usage report.

vCPU. Virtual central processing unit. The standard unit of compute capacity allocated to a cloud instance, representing a share of a physical processor core.