▸ Independent Standard · v1.0 · April 21, 2026

CEH™
(Compute Energy Hour)

The independent, auditable standard unit for measuring how efficiently energy is converted into compute output — across any hardware, any facility, any energy source. Originated by Oak Ridge Management / DGE&I.

View CEH™ Index → Methodology Download Whitepaper

▸ CEH™ Formula · v1.0

CEH™ =
  ( TDP_W × Units × Util% × PUE )
  ÷ 1,000
  ÷ Output_Units_per_Hour

TDP_W Thermal Design Power per unit (watts)

Units Count of accelerator units in configuration

Util% Average utilization rate (0–1)

PUE Power Usage Effectiveness (facility ratio)

Output Compute Output Units per hour (COU-dependent)

What CEH™ Measures

CEH™ Rate

The core measure: kilowatt-hours of energy consumed per unit of compute output. Lower is more efficient. Energy-agnostic — the formula is identical regardless of energy source.

kWh / output_unit

CEH™ Cost

The economic dimension: CEH™ Rate multiplied by the energy rate ($/kWh). Enables direct comparison across grid-connected and BTM infrastructure using the same energy source rate input.

CEH™ × $/kWh (any source)

CEH™ Carbon

The environmental dimension: CEH™ Rate multiplied by the grid or facility carbon intensity (kg CO₂/kWh). Comparable across regions and energy mixes without conflating efficiency with source attribution.

CEH™ × kg CO₂/kWh

Compute Output Unit (COU) Definitions

The COU is the workload-specific unit of output that defines the denominator in the CEH™ formula. COU selection is workload-portable: CEH™ rates are comparable within a workload class. Cross-class comparisons require normalized COU translation per the v1.0 methodology.

LLM Inference

Tokens / hour

Output tokens/sec · Llama-class · batch 8 · vLLM

Primary workload class for CEH™ Index v1.0. Throughput sourced from MLPerf v5.1, CUDO, Koyeb, Spheron (2025–26).

Model Training

TFLOPS / hour

Sustained FP16/BF16 floating-point ops/sec

Mixed-precision training workloads. Equivalent CEH™ thresholds defined in v1.0 methodology specification.

Rendering / VFX

Frames / hour

GPU rasterization + ray tracing pipeline

Studio-grade rendering workloads. Grade thresholds scale to normalized frame complexity factor.

HPC / Simulation

Iterations / hour

Simulation steps/sec · domain-specific normalization

Scientific computing and simulation workloads. Requires per-study COU normalization per methodology spec.

COU definitions are public and workload-portable. Cross-class comparison requires normalized COU translation as specified in CEH™ v1.0 methodology. The CEH™ Index v1.0 benchmarks LLM Inference (Tokens) exclusively.

CEH™ Efficiency Grade Scale

< 2.0 × 10⁻¹⁰ kWh/token

Deploy as primary inference hardware

2.0 – 4.0 × 10⁻¹⁰

Cost-competitive; preferred configuration

4.0 – 8.0 × 10⁻¹⁰

Viable for mixed workloads; monitor utilization

8.0 – 15.0 × 10⁻¹⁰

Suitable for low-priority or burst workloads

15.0 – 30.0 × 10⁻¹⁰

Avoid for sustained production

> 30.0 × 10⁻¹⁰

Retire or limit to legacy use only

Thresholds calibrated to LLM Inference (token output). Equivalent thresholds for training (TFLOPS), rendering (frames), and HPC (iterations) are defined in the CEH™ v1.0 methodology specification.

Standard Documentation

CEH™ Index

Benchmark Table

Full 10-configuration dynamic benchmark table with grade, cost, and carbon columns.

View Index →

CEH™ Calculator

Financial Model

Interactive BTM vs. grid CEH™ cost model. Input your configuration, output the spread.

Open Calculator →

Methodology

Disclosure

Full formula derivation, data sources, version archive, and replication protocol.

CEH™(Compute Energy Hour)

CEH™ Rate

CEH™ Cost

CEH™ Carbon

CEH™
(Compute Energy Hour)