▸ CEH™ Standard · Methodology

Methodology & Derivation

The complete derivation framework, Compute Output Unit definitions, certification criteria, and versioning protocol for the CEH™ Standard.

▸ Formal Derivation

The Five-Step CEH™ Calculation

1
Raw Power Draw
Multiply hardware TDP (Watts) by number of units to establish the nominal power envelope of the configuration.
Raw Power = TDP_W × Units
2
Effective Power
Apply utilization rate — the measured average percentage of peak TDP consumed during the workload benchmark window. Must be measured, not assumed.
Effective Power = Raw Power × Utilization%
3
Facility Power (PUE)
Apply PUE (Power Usage Effectiveness) to account for cooling, power conversion, and infrastructure overhead. Hyperscale: 1.1–1.2. Co-location: 1.4–1.7. Purpose-built BTM: 1.05–1.1.
Facility Power = Effective Power × PUE
4
Convert to kWh
Divide total watts by 1,000 to yield kilowatt-hours per hour of operation at the declared configuration and utilization.
kWh/hr = Facility Power ÷ 1,000
5
Compute CEH™
Divide kWh by compute output units delivered during that hour. Output unit is standardized per the COU definitions below. Result is expressed in kWh per compute unit-hour.
CEH™ = kWh/hr ÷ Output_Units_per_Hour
▸ Compute Output Unit (COU) Definitions · v1.0
Workload Type COU Definition Unit Label Notes
LLM Inference Output tokens generated Tokens Batch size 8, standard seq length, vLLM
Model Training FP16/FP8 TFLOPS sustained TFLOPS End-to-end training throughput
Image Rendering Completed frames rendered Frames At target resolution and quality setting
Scientific HPC Simulation iterations completed Iterations Problem-specific; must be declared in benchmark disclosure
Custom / General Standardized Compute Unit (SCU) SCU User-defined; full methodology must be published
▸ Certification Disclosure Requirements

A CEH™-certified benchmark must disclose all of the following parameters to be considered reproducible and comparable under the v1.0 standard:

  • ✓Hardware configuration: chip model, unit count, TDP (per manufacturer spec or measured)
  • ✓Utilization rate: measured average during benchmark window, not theoretical peak
  • ✓PUE: declared facility PUE or measured PUE for the specific deployment
  • ✓Workload type and COU definition: must match a v1.0 standard COU or declare a custom SCU with full methodology
  • ✓Throughput measurement framework: vLLM, TensorRT-LLM, etc.; batch size; sequence length
  • ✓Energy rate used for CEH™ Cost calculation, with source and contract type noted
  • ✓Carbon intensity source and vintage for CEH™ Carbon calculation
▸ Versioning

This document constitutes CEH™ Standard v1.0. The standard is versioned semantically. Version increments are triggered by: expansion of COU definitions to new workload types, revision to benchmark methodology, or material changes to the derivation formula. All versions are archived and dated. Comments on methodology may be submitted via the contact form.