inference/research
verifiedhigh · 0.85updated 2026-10-05

LBNL 2014 US PUE by space type - midtier: 1.9 (component sum: 1 | 0.05 | 0.2 | 0.63 | 0.02 -> IT 1, transformer 0.05, UPS 0.20, cooling 0.63, lighting 0.02)

1.9 dimensionless PUE ratio (total facility energy / IT equipment energy), 2014 baseline. Not an accuracy score and not a speedup: 1.0 means the facility overhead is nil.

Fields

id
benchmarks/pue-lbnl-2016-us-midtier-2014-1-9
Metric
dimensionless_ratio
Value
1.9
Unit
dimensionless PUE ratio (total facility energy / IT equipment energy), 2014 baseline. Not an accuracy score and not a speedup: 1.0 means the facility overhead is nil.
Methodology
MODELED BASELINE, TABLE 4 ('2014 PUE by Space Type') of the DOE/BL United States Data Center Energy Usage Report. The row decomposes total PUE into IT (=1 by definition), transformer, UPS, cooling and lighting contributions, and the components sum to the stated total - which is how the table was verified here rather than trusted. Space types are IDC-defined and span both internal enterprise spaces and service-provider spaces; the report states service provider data centers are INCLUSIVE of colocation facilities. This is the most granular public per-facility-type PUE breakdown available from an authoritative US government source. engine_id is ollama only as a schema placeholder: facility metric, no engine.
Measured by
third_party
Reproducible
yes
notes
Midtier: 1.9. The report groups service-provider spaces (which it states are INCLUSIVE of colocation facilities) across the same five size categories as internal spaces plus hyperscale. This is the closest thing to a published colocation PUE in the DOE/BL public record, and it is materially worse than hyperscale - relevant because colocation is where most third-party inference actually runs. CAVEAT ON AGE: these are 2014 values and the 2016 report frames the 2020 trajectory separately (dynamic range, consolidation). For a current planning number prefer the 2025 Update's 1.45 national all-equipment average. The 2014 table is retained because the modern public sources publish only fleet or national AVERAGES - this table is the only public breakdown that shows a colocation-class facility is closer to 1.9 than to 1.2. METRIC CORRECTION 2026-10-05 (schema authority pass, agent w5-schema). Value UNCHANGED; `metric` moved quality -> dimensionless_ratio. PUE is a ratio by DEFINITION (total facility energy / IT equipment energy) and carries no faster/slower reading, so it is a dimensionless_ratio rather than a speedup_ratio. It was previously filed as `quality`, the catch-all that gotcha bench-metric-enum-hides-a-dimensionless-ratio names as the reason `quality` cannot be trusted as a semantic filter: that key was spanning accuracy deltas in percentage points, PUE ratios and power ratios simultaneously.

Notes

Midtier: 1.9. The report groups service-provider spaces (which it states are INCLUSIVE of colocation facilities) across the same five size categories as internal spaces plus hyperscale. This is the closest thing to a published colocation PUE in the DOE/BL public record, and it is materially worse than hyperscale - relevant because colocation is where most third-party inference actually runs. CAVEAT ON AGE: these are 2014 values and the 2016 report frames the 2020 trajectory separately (dynamic range, consolidation). For a current planning number prefer the 2025 Update's 1.45 national all-equipment average. The 2014 table is retained because the modern public sources publish only fleet or national AVERAGES - this table is the only public breakdown that shows a colocation-class facility is closer to 1.9 than to 1.2.

METRIC CORRECTION 2026-10-05 (schema authority pass, agent w5-schema). Value UNCHANGED; `metric` moved quality -> dimensionless_ratio. PUE is a ratio by DEFINITION (total facility energy / IT equipment energy) and carries no faster/slower reading, so it is a dimensionless_ratio rather than a speedup_ratio. It was previously filed as `quality`, the catch-all that gotcha bench-metric-enum-hides-a-dimensionless-ratio names as the reason `quality` cannot be trusted as a semantic filter: that key was spanning accuracy deltas in percentage points, PUE ratios and power ratios simultaneously.

Cross-references

Engine (1)