What is the PUE formula?

PUE = Total Facility Power / IT Equipment Power (ISO/IEC 30134-2). Total facility power includes IT load plus cooling, UPS losses, transformer losses, PDU losses, lighting, security, and fire suppression. DCiE = 1/PUE.

How does cooling type affect PUE?

Cooling overhead in the PUE model equals IT load × a type-specific fraction representing the inverse of COP: CRAC ~0.45 (COP 2.8), CRAH ~0.38 (COP 4.0), in-row ~0.32 (COP 5.0), rear-door ~0.22 (COP 8.0), direct liquid ~0.12 (COP 15), immersion ~0.08 (COP 25). Lower COP = higher cooling overhead = higher PUE.

What is the global average PUE benchmark?

The Uptime Institute 2024 Global Survey reports the worldwide average at 1.58. Best-in-class hyperscale facilities (Google, Meta, AWS) achieve 1.10–1.20. Enterprise colocations typically fall in the 1.40–1.60 range. Legacy facilities often exceed 2.0.

PUE Calculator — Technical Manual & Methodology

01 Purpose & engineering basis

The calculator quantifies the Power Usage Effectiveness (PUE) of a data center from a bottom-up facility power decomposition. It answers three design questions: what fraction of drawn power reaches IT equipment, how does cooling architecture drive overhead, and what is the annual energy cost and carbon footprint.

PUE is defined in ISO/IEC 30134-2:2016 (adopted by The Green Grid as the industry standard) and measured per ASHRAE TC 9.9 operating envelope guidance. The tool models PUE from first principles rather than from a lookup table, so it can predict the impact of architecture changes before they are built. The global benchmark is 1.58 (Uptime Institute Global Survey 2024); hyperscale leaders achieve 1.10–1.20.

The governing identity is:

PUE = P_total / P_IT DCiE = 1 / PUE (Data Center Infrastructure Efficiency) ISO/IEC 30134-2 §5. P_IT is the IT equipment load (denominator). P_total is the sum of all loads (numerator).ISO/IEC 30134-2 · STANDARD

02 Inputs

Fourteen inputs across five groups — IT infrastructure, cooling system, power distribution, facility loads, and PRO advanced parameters. Ranges are enforced client-side; defaults reflect a typical Tier III colocation at 1 MW IT load.

Field	ID	Unit	Default	Tier	Meaning
IT Load	`itLoad`	kW	1 000	Free	Total IT equipment power — the PUE denominator. Range 50–100 000 kW.
Rack Density	`rackDensity`	kW/rack	8	Free	Used to auto-calculate rack count = ceil(IT load / density). Options: 5, 8, 15, 30.
Cooling Type	`coolingType`	enum	crah	Free	Sets the cooling overhead fraction (proxy for inverse COP). See §3.
Containment	`containment`	enum	cold	Free	Airflow containment strategy reduces cooling overhead by a fixed fraction. See §3.
Climate Zone	`climateZone`	enum	subtropical	Free	Multiplies the cooling overhead (tropical-humid = 1.15, cold = 0.78). See constants §4.
UPS Type	`upsType`	enum	double	Free	Determines the load-dependent UPS efficiency curve. See §3.
UPS Load Factor	`upsLoadFactor`	%	50	Free	% of rated UPS capacity utilised. Governs which point on the efficiency curve is used.
Redundancy	`redundancy`	enum	n1	Free	Scales UPS losses by redundancy overhead multiplier (N=1.00, N+1=1.10, 2N=1.25, 2N+1=1.35).
Lighting	`lighting`	W/m²	8	Free	Lighting power density. Floor area = rack count × 3 m²/rack. Lighting kW = W/m² × area / 1000.
Security	`security`	kW/rack	0.5	Free	Physical security load (CCTV, access control, NVR). Security kW = rate × rack count.
Fire Suppression	`fire`	kW/rack	0.1	Free	Detection + suppression standby power. Fire kW = rate × rack count.
Economizer Mode	`economizer`	enum	none	PRO	Reduces cooling kW by 15% (air-side) or 25% (water-side). Driven by free-cooling hours in climate data.
Supply Air Temp	`supplyTemp`	°C	20	PRO	Each °C above 20 saves 3% cooling energy (ASHRAE TC 9.9 range 18–27°C).
Transformer Loss	`transformerLoss`	%	1.5	PRO	Voltage-transformation losses. Transformer kW = IT load × tx%.
PDU Type	`pduType`	enum	metered	PRO	PDU efficiency: basic 98%, metered 98.5%, monitored/switched 99%. PDU loss = IT × (1 − eff).
Energy Cost	`energyCost`	$/kWh	0.10	PRO	Blended electricity rate for annual cost calculation. Compare with `DATA.regions[*].powerKwh`.

03 Calculation methodology

The calculator decomposes total facility power into six additive components, then applies PUE = P_total / P_IT. All intermediate values are deterministic given the inputs.

Step 1 — Rack count (derived)

rackCount = ceil( itLoad / rackDensity ) Used to scale facility loads (lighting area, security kW, fire kW). Auto-shown in the UI; can be overridden.

Step 2 — Cooling power

P_cool = itLoad × ( coolingOverhead[type] × climateMult[zone] − containmentReduction[containment] ) if econMode === 'airside': P_cool × 0.85 if econMode === 'waterside': P_cool × 0.75 if PRO supplyTemp ≠ 20°C: P_cool × ( 1 − (supplyTemp − 20) × 0.03 ) P_cool = max( P_cool, itLoad × 0.02 ) (floor: 2% of IT) coolingOverhead is a type-keyed fraction approximating 1/COP × overhead share (see constants §4). climateMult adjusts for ambient cooling penalty. Containment reduces mixing losses. Economizer applies free cooling. Each °C above 20°C cuts cooling 3% via higher chiller efficiency and extended economizer hours (ASHRAE TC 9.9).ASHRAE TC 9.9 · STANDARD

Step 3 — UPS losses

upsEff = interpolate( upsEfficiency[type], upsLoadFactor ) (linear between 25/50/75/100%) P_ups = itLoad × ( 1/upsEff − 1 ) × redundancyMult[redund] Efficiency is load-dependent (curves per type, §4). Double-conversion at 50% load = 92% → loss = 8.7%. Redundancy overhead multiplier scales losses for standby capacity.IEEE 1100 / UPS vendor data

Step 4 — Transformer & PDU losses (PRO)

P_tx = itLoad × txLoss (default 1.5%) P_pdu = itLoad × ( 1 − pduEfficiency[type] ) (metered: 0.5%) Transformer losses are load-proportional in this model (conservative; actual no-load losses are constant). DOE 2016-compliant dry-type transformers achieve 0.5–1.5% loss vs legacy 1.5–2.5%.DOE 10 CFR Part 431 / IEEE C57.12

Step 5 — Facility loads

floorArea = rackCount × 3 (m², 3 m²/rack rule-of-thumb) P_lighting = lightingWm2 × floorArea / 1000 P_security = secKwRack × rackCount P_fire = fireKwRack × rackCount P_facility = P_lighting + P_security + P_fire Lighting, security, and fire suppression are minor but measurable components of facility overhead. Lighting typically 1–3% of total; security and fire together <1%.

Step 6 — PUE, DCiE, energy, carbon

P_total = P_IT + P_cool + P_ups + P_tx + P_pdu + P_facility PUE = P_total / P_IT ← RZEngine.models.pue.pueFromInputs(itLoad, P_total) DCiE = 1 / PUE ← RZEngine.models.pue.dcie(PUE) annualEnergy = P_total × 8 760 [kWh/yr] annualCost = annualEnergy × energyCostPerKWh CO2_tyr = annualEnergy × gridFactor / 1 000 [t CO₂e/yr] (PRO; default 0.5 kg/kWh) Engine functions pueFromInputs and dcie are in RZEngine.models.pue (rz-engine.js §S4). Grid carbon factor defaults to 0.5 kg CO₂/kWh blended; pro mode uses DATA.carbon.gridFactor[region] per climate/region.ISO/IEC 30134-2 · STANDARD; DATA.carbon.gridFactor · rz-engine.js

Partial-load PUE curve (free panel)

PUE(l) = 1 + overhead × ( 0.55/l + 0.45 ) where overhead = designPUE − 1, l = IT utilisation fraction ∈ [0.05, 1.00] Fixed-share 0.55 means 55% of facility overhead is invariant to IT load (cooling base load, transformer no-load, lighting, security) — it doesn't scale down with IT, so PUE rises at low utilisation. Powered by RZEngine.models.pue.partialLoadPUE(designPUE, loadFraction).Green Grid partial-load PUE guidance; DATA.sources['pue.partialLoad']

04 Constants & data sources

Every constant is authored in the calculator JS or in the deep-frozen DATA block of rz-engine.js (v2.0), with a DATA.sources citation. Key tables:

Cooling Type	COP (nominal)	Overhead fraction	Containment reductions
CRAC (Computer Room AC)	~2.8	`0.45`	None: 0.00 · Hot-aisle: 0.10 · Cold-aisle: 0.12 · Chimney: 0.15
CRAH (Computer Room Air Handler)	~4.0	`0.38`
In-Row Cooling	~5.0	`0.32`
Rear-Door Heat Exchanger	~8.0	`0.22`
Direct Liquid Cooling (DLC)	~15	`0.12`
Immersion Cooling	~25	`0.08`

Climate Zone	Climate Multiplier	Free-Cooling Hrs/yr	Example cities
Tropical Humid	`1.15`	0	Jakarta, Singapore
Tropical Dry	`1.08`	500	Dubai, Phoenix
Subtropical	`1.00`	1 500	Hong Kong, Sydney
Temperate	`0.92`	3 000	London, San Francisco
Continental	`0.85`	4 500	Chicago, Frankfurt
Cold	`0.78`	6 000	Stockholm, Helsinki

UPS Type	Eff @ 25%	Eff @ 50%	Eff @ 75%	Eff @ 100%
Double-Conversion (Online)	88%	92%	94%	95%
Line-Interactive	93%	96%	97%	97.5%
Rotary / DRUPS	90%	94%	96%	97%
Flywheel Hybrid	95%	97%	98%	98.5%

Region	Power cost ($/kWh)	Grid factor (kg CO₂/kWh)	Source
United States	`0.090`	`0.37`	EIA industrial + DC PPA blend
Europe (blended)	`0.235`	`0.23`	Eurostat non-household electricity
Asia-Pacific (blended)	`0.110`	`0.55`	IEA / regional utility filings
Indonesia (ID)	`0.075`	`0.68`	PLN tariff / IEA 2025
Singapore (SG)	`0.180`	`0.41`	EMA Singapore 2025
Malaysia (MY)	`0.070`	`0.55`	TNB / IEA 2025

PUE Benchmark	Cooling	Tier II	Tier III	Tier IV	Source
Air-cooled	CRAH / CRAC	`1.62`	`1.50`	`1.44`	Uptime Institute Global Survey 2026
In-row	close-coupled	`1.34`	`1.27`	`1.22`	Uptime Institute 2026 close-coupled cohort
Rear-door HX	RDHX	`1.24`	`1.18`	`1.14`	Uptime Institute 2026 rear-door cohort
Liquid / DLC	direct-to-chip	`1.22`	`1.15`	`1.10`	Uptime Institute 2026 / ASHRAE TC9.9
Immersion	single/two-phase	`1.07`	`1.04`	`1.03`	Uptime Institute 2026

Benchmark PUE 1.58: Uptime Institute 2024 Global Survey global average. DATA.sources['pueDefaults'] and DATA.sources['pueMatrix'] in rz-engine.js cite Uptime Global PUE Survey 2026 as of the last engine refresh.

05 Outputs

Output	Formula	Unit	Tier	Interpretation
PUE	`P_total / P_IT`	ratio	Free	≤1.20 = Excellent; 1.20–1.40 = Very Good; 1.40–1.60 = Good; 1.60–1.80 = Average; >2.0 = Poor.
DCiE	`1 / PUE × 100%`	%	Free	Percentage of total power that reaches IT. DCiE 64.5% means 35.5% is non-IT overhead.
Total Facility Power	`sum of all components`	kW	Free	Numerator of PUE. Capacity planning input (utility connection sizing).
Cooling Load	`P_cool`	kW	Free	Largest non-IT load. Use to size chiller plant and cooling-tower capacity.
UPS Loss	`P_ups`	kW	Free	AC-DC-AC conversion loss. Shrinks with higher-efficiency topology or higher load factor.
Annual Energy	`P_total × 8 760`	MWh / GWh	Free	Total facility electricity consumption per year (8 760 h/yr = `DATA.hoursPerYear`).
Annual Cost	`annualEnergy × $/kWh`	USD	PRO	Estimated annual electricity spend (excludes demand charges and ToU variations).
CO2/Year	`annualEnergy × gridFactor / 1 000`	t CO₂e	PRO	Scope 2 emissions. Uses 0.5 kg/kWh default blended or `DATA.carbon.gridFactor[region]`.
Free Cooling Hours	climate lookup	hrs/yr	PRO	Estimated annual hours when ambient enables economizer operation. From `climateData[zone].freeCoolingHrs`.
Partial-load PUE	`RZEngine.models.pue.partialLoadPUE(PUE, 0.50)`	ratio	Free panel	PUE at 50% IT utilisation — shows efficiency degradation at partial load.
WUE	`DATA.water.wueByType[cooling]`	L/kWh	Free panel	Water Use Effectiveness companion: air 1.80, rear-door 1.10, DLC 0.50, immersion 0.10 L/kWh.
Annual Water	`WUE × IT_kWh / 1 000`	m³/yr	Free panel	Total water consumption for cooling per year.

06 Worked example

1 000 kW IT load (default preset), CRAH cooling, cold-aisle containment, subtropical climate, double-conversion UPS at 50% load factor, N+1 redundancy, standard lighting (8 W/m²), standard security (0.5 kW/rack), FM-200 fire (0.1 kW/rack). PRO defaults: no economizer, supply temp 20°C, transformer 1.5%, metered PDU, energy cost $0.10/kWh. All values computed from the documented constants — no lookup-only numbers.

Rack count: ceil(1 000 / 8) = 125 racks
Cooling (CRAH, cold-aisle, subtropical):
1 000 × (0.38 × 1.00 − 0.12) = 1 000 × 0.26 = 260 kW
UPS efficiency (double-conversion, 50% load): interpolate curve → 92%
P_ups = 1 000 × (1/0.92 − 1) × 1.10 = 1 000 × 0.0870 × 1.10 = 95.7 kW
Transformer losses (PRO default 1.5%): 1 000 × 0.015 = 15 kW
PDU losses (metered, 98.5% eff): 1 000 × (1 − 0.985) = 15 kW
Floor area: 125 × 3 = 375 m²
Lighting: 8 × 375 / 1 000 = 3.0 kW
Security: 0.5 × 125 = 62.5 kW
Fire: 0.1 × 125 = 12.5 kW
P_facility = 78.0 kW
Total facility power:
1 000 + 260 + 95.7 + 15 + 15 + 78 = 1 463.7 kW
PUE: 1 463.7 / 1 000 = 1.464 → rating: A — Very Good
DCiE: 1 / 1.464 × 100 = 68.3%
Annual energy: 1 463.7 × 8 760 = 12 822 MWh/yr (12.8 GWh)
Annual cost (PRO, $0.10/kWh): 12 822 000 × 0.10 = $1.28M/yr
CO2 (PRO, blended 0.5 kg/kWh): 12 822 000 × 0.5 / 1 000 = 6 411 t CO₂e/yr
Partial-load PUE at 50% IT:
overhead = 1.464 − 1 = 0.464 · PUE(0.50) = 1 + 0.464 × (0.55/0.50 + 0.45) = 1 + 0.464 × 1.55 = 1.719
WUE (CRAH → air key): DATA.water.wueByType.air = 1.80 L/kWh
Annual water: 1.80 × 1 000 × 8 760 / 1 000 = 15 768 m³/yr

Engineering reading: at this configuration the 260 kW cooling load is the dominant overhead (17.8% of total), followed by security (4.3%), UPS losses (6.5%), and PDU/transformer (2% each). Upgrading from CRAH to in-row cooling reduces the overhead fraction from 0.38 to 0.32 and would drop PUE by roughly 0.06. Adding a water-side economizer in subtropical conditions (1 500 hrs/yr) saves an additional 25% of cooling or ~65 kW on this load, translating to ~0.065 PUE reduction.

07 References & standards

ISO/IEC 30134-2:2016 — Information technology — Data centres key performance indicators — Part 2: Power usage effectiveness (PUE). Definitive normative basis for the PUE formula and measurement boundary.
Green Grid PUE White Papers — WP#6 (PUE: A comprehensive examination of the metric), WP#49 (Measuring & managing data center energy use), WP#66 (PUE measurement granularity L1–L3). Source of the L1/L2/L3 measurement levels implemented in the PRO Seasonal panel.
ASHRAE TC 9.9 (2021 ed.) — Thermal Guidelines for Data Processing Environments. Basis for supply air temperature range (18–27°C, A1 class), each-degree-Celsius cooling savings, and ASHRAE class definitions A1–A4.
Uptime Institute Global Survey 2024 / 2026 — Primary benchmark for global average PUE 1.58, tier x cooling PUE matrix, and the Tier I–IV availability / redundancy standard. Cited in DATA.sources['pueDefaults'] and DATA.sources['pueMatrix'] in rz-engine.js.
IEEE 1100 (Emerald Book) — Recommended practice for powering and grounding electronic equipment. Basis for UPS efficiency topology descriptions and load-curve behavior.
DOE 10 CFR Part 431 — Energy Conservation Standards for Distribution Transformers (DOE 2016 rule). Basis for transformer loss ranges (0.5–2.5%).
GHG Protocol Corporate Standard — Scope 2 grid emission factors methodology. DATA.carbon.gridFactor values and source citations in rz-engine.js §DATA.sources.
Engine source — rz-engine.js (RZEngine v2.0): DATA.pueDefaults, DATA.pueMatrix, DATA.coolingTypes, DATA.water.wueByType, DATA.carbon.gridFactor, models.pue.*. Calculator: pue-calculator.html.

08 Assumptions & limitations

The calculation assumes a single thermodynamic zone at steady-state load; seasonal variation is modelled only in the PRO monthly chart (not in the base PUE output). Cooling overhead fractions are calibrated to typical-installation COP for each technology type; site-specific COP varies with supply water temperature, ambient wet-bulb, and part-load conditions. The 3 m²/rack floor area used for lighting calculations is a rule-of-thumb; dense hot-aisle configurations may pack more tightly. Redundancy multipliers scale UPS loss proportionally — this is conservative for 2N systems where one UPS string is entirely bypassed at low load.

Transformer losses are modelled as load-proportional (a slight overstatement for DOE 2016-compliant units that have low constant core losses). Annual hours of 8 760 is the DATA.hoursPerYear constant; leap-year and planned-downtime corrections are not applied. Energy cost uses a flat $/kWh blended rate; demand charges (typically 20–30% of actual bills) are not included.

This manual and the calculator are an engineering education and pre-design aid. PUE results are screening estimates; final energy and carbon reporting requires metered measurements per Green Grid L2/L3 methodology and validation by a qualified data center energy engineer.