What is PUE? Data center power usage effectiveness explained

What is PUE? It is the standard metric for how efficiently a data center uses electricity. PUE stands for power usage effectiveness, and the formula is PUE = total facility energy ÷ IT equipment energy. The numerator is all the energy the facility consumed over a period; the denominator is the part of it that actually reached servers, storage and network equipment. For example, if a facility used 8,760,000 kWh in a year and its IT equipment used 6,000,000 kWh of that, PUE = 8,760,000 ÷ 6,000,000 = 1.46: for every kWh the IT equipment consumed, the facility spent another 0.46 kWh delivering power to it, cooling it and lighting the room. The Green Grid introduced PUE in 2007, and the international standard ISO/IEC 30134-2 now defines it the same way. It is the most widely quoted data center energy metric and the number regulators and customers increasingly ask for, but it has clear limits. The sections below cover what goes into it, how to read a value, why lower is better and what it cannot tell you.

What counts in the numerator and the denominator

Split a facility’s energy by where it goes and the two sides of the ratio become clear. Take a small facility with a 100 kW IT load; at a given moment its power draw breaks down like this (illustrative figures):

Where the power goes	Power	Counted in
IT equipment: servers, storage, switches, routers	100 kW	Numerator and denominator
Cooling: compressors, indoor and outdoor fans, pumps, cooling towers	38 kW	Numerator only
Power distribution losses: transformers, UPS, switchgear and cabling	8 kW	Numerator only
Lighting, security, environmental monitoring, access control	4 kW	Numerator only
Total	150 kW	PUE = 150 ÷ 100 = 1.50

Think of the numerator as everything the facility spends to keep running, and the denominator as what the working equipment finally receives; the difference is infrastructure overhead. In practice you divide energy (kWh) over the same period rather than instantaneous power, and a figure you publish should cover a full year.

A few loads are easy to put on the wrong side. The rule is whether the energy is consumed before or after the power input of the IT equipment:

• Fans inside servers and power supply conversion losses: IT. They sit behind the server’s power inlet, so no meter can separate them out. This comes up again in the limitations below.
• KVM switches and out-of-band management switches in the racks: IT, because they are IT equipment themselves.
• In-row coolers, fans in aisle containment, and coolant distribution units (CDUs) for liquid cooling: cooling, so non-IT, even though they sit in the rack rows.
• Environmental monitoring controllers, access control and cameras: non-IT, even where they are fed from the UPS output.
• Losses in rack PDUs and branch circuits: depend on where IT energy is metered. Metered at the UPS output, these losses land in the denominator and PUE comes out slightly low; metering at PDU outputs or at the equipment inputs gets closer to the true IT energy.

What a PUE value means: 1.2, 1.5 and 2.0

There are two intuitive ways to read a PUE value: how much extra energy each kWh of IT energy costs, and what share of total energy never reaches IT. That share is (PUE − 1) ÷ PUE:

PUE	Extra energy per 1 kWh of IT energy	Non-IT share of total energy	DCiE (= 1 ÷ PUE)
1.1	0.1 kWh	9.1%	90.9%
1.2	0.2 kWh	16.7%	83.3%
1.3	0.3 kWh	23.1%	76.9%
1.5	0.5 kWh	33.3%	66.7%
1.8	0.8 kWh	44.4%	55.6%
2.0	1.0 kWh	50.0%	50.0%
2.5	1.5 kWh	60.0%	40.0%

So at PUE 2.0, half of the electricity bill is not spent on servers at all; at PUE 1.2 the share drops to one sixth. The last column, DCiE (data center infrastructure efficiency), is the reciprocal of PUE expressed as a percentage. It was common in the early years; industry reporting and regulation now use PUE almost exclusively.

Some reference points: 1.0 is the theoretical floor, meaning zero energy for cooling, distribution losses and lighting, which no real facility reaches. Regulators increasingly look at the number: the EU Energy Efficiency Directive requires data centers with 500 kW or more of installed IT power to report PUE along with other indicators, and some countries set maximum PUE values for new facilities. Whether a particular figure is reasonable also depends on the facility’s size, climate and occupancy, so the number on its own says less than it seems.

Why a PUE closer to 1 is better

The reason is direct: for the same IT load, a lower PUE means less total energy. Take a hypothetical facility with an average IT load of 400 kW, running 8,760 hours a year, for 3,504,000 kWh of IT energy:

	PUE 1.60	PUE 1.35
Annual total energy	5,606,400 kWh	4,730,400 kWh
of which non-IT	2,102,400 kWh	1,226,400 kWh
Annual power bill (assumed $0.10/kWh, for illustration only)	about $561,000	about $473,000

With the same servers, going from PUE 1.60 to 1.35 saves 876,000 kWh and about $87,600 a year at that assumed tariff. For a colocation provider that sells racks with power included, the saving goes straight to margin; in a facility whose utility feed is already fully subscribed, a lower PUE also means the same incoming capacity can carry more IT load.

The further down you go, though, the harder each step gets. With the IT load unchanged, going from P1 to P2 cuts total energy by (P1 − P2) ÷ P1:

Reduction	Total energy saved	Typically achieved by
2.0 → 1.5	25.0%	Blanking panels, a higher supply air temperature, switching off surplus cooling units and humidification
1.5 → 1.3	13.3%	Hot or cold aisle containment, variable-speed fans and pumps, a higher UPS load factor
1.3 → 1.2	7.7%	Extensive free cooling, indirect evaporative cooling or liquid cooling, which need substantial investment

The first steps are mostly operational fixes and small retrofits; the last ones depend on the cooling architecture itself. That is why existing facilities usually aim for somewhere around 1.5, while figures like 1.2 mostly come from new, large data centers.

What PUE does not tell you: five limitations

PUE answers one question: how much extra energy does the infrastructure use to deliver power to IT? It cannot show anything beyond that, and sometimes it points the wrong way.

1. It ignores how much work the IT equipment does

A server idling at 5% utilization and one running flat out count the same in the denominator. Suppose a facility has a 100 kW IT load and its non-IT power is 20 kW of fixed overhead plus 30% of the IT load, for a total of 150 kW and a PUE of 1.50. Now decommission or consolidate 25 kW of servers that have been idle for months:

	Before	After
IT power	100 kW	75 kW
Non-IT power	20 + 30 = 50 kW	20 + 22.5 = 42.5 kW
Total power	150 kW	117.5 kW
PUE	1.50	about 1.57

Total power falls by 21.7%, yet PUE gets worse, because the fixed overhead is now spread over a smaller IT load. A team judged on PUE alone has no reason to do this kind of genuinely energy-saving work.

2. Server fans count as IT, so PUE can improve on paper

Raising the supply air temperature lets the cooling plant use less energy, but warmer inlet air makes server fans spin faster, and that extra energy lands in the denominator. Suppose that in the same facility cooling drops by 10 kW and server fans add 4 kW: IT power becomes 104 kW, total power 150 − 10 + 4 = 144 kW, and PUE falls from 1.50 to about 1.38, an improvement of 0.12, while total power fell by only 6 kW, or 4%. PUE improves far more than real energy use, so when you tune temperatures, watch total energy as well.

To address this, researchers proposed ITUE and TUE in 2013. ITUE = total IT equipment energy ÷ energy used by the compute components (CPU, memory, storage and so on), which captures the overhead of fans and power conversion inside the servers; TUE = ITUE × PUE, the efficiency from the utility feed all the way to the compute components. Both need component-level data from inside the servers, so they are mostly used in research and in large self-built data centers.

3. It leaves out water and carbon

Evaporative cooling and cooling towers can lower PUE noticeably, at the cost of water. Two facilities with the same PUE, one running on coal power and one on wind and solar, have very different carbon emissions. For these, look at WUE (water usage effectiveness), CUE (carbon usage effectiveness) and the renewable energy share.

4. Heat reuse does not show up

Piping server waste heat into district heating or neighboring buildings saves real energy, but neither side of the PUE ratio changes. The metric for this is ERE (energy reuse effectiveness): ERE = (total energy − reused energy) ÷ IT energy, which can fall below 1. A related metric, ERF (energy reuse factor), is reused energy ÷ total energy. By definition PUE cannot be lower than 1; if you see a claim of a PUE below 1, it almost certainly credits reused energy and should properly be called ERE.

5. PUE values from different sites are not directly comparable

A PUE of 1.4 means different things in a cool northern climate and in a hot, humid one; at 30% and at 80% occupancy; with 2N and with N+1 power; metered at the UPS output or at PDU outputs; as an annual average or a winter-only figure. Before comparing, line up five things: climate, load level, redundancy, metering point and reporting period. The same applies to published numbers: a design PUE, one month’s measured value and a full-year measured value are not interchangeable.

Other efficiency metrics to read alongside PUE

Metric	Formula	What it adds to PUE
Cooling load factor, power load factor	Cooling energy ÷ IT energy; power distribution losses ÷ IT energy	Breaks PUE down to show whether cooling or power losses dominate
pPUE (partial PUE)	All energy within a zone ÷ IT energy within that zone	Assesses one module or hall on a campus with a shared chiller plant
WUE	Annual water use (L) ÷ annual IT energy (kWh)	Water
CUE	Annual emissions (kgCO₂e) ÷ annual IT energy (kWh)	Carbon, which depends on the energy mix
Renewable energy share	Renewable energy ÷ total energy	Share of green power
ERE, ERF	(Total energy − reused energy) ÷ IT energy; reused energy ÷ total energy	Heat reuse
ITUE, TUE	Total IT energy ÷ compute component energy; ITUE × PUE	Fans and power conversion losses inside servers
Rack occupancy	Racks in use ÷ racks available	How load level affects PUE, and whether capacity sits idle

For day-to-day management the most useful combination is PUE for the infrastructure, the two load factors to locate where the overhead comes from, and rack occupancy plus server utilization to spot waste on the IT side. Read together, they keep any single number from misleading you.

Tracking PUE inputs and idle servers in DCIM

Calculating PUE takes two sets of data: the numerator comes from the main meters in the power distribution or facility monitoring system, and the denominator has to be totaled from UPS units, remote power panels or rack PDUs. Rack-level data is DCIM territory. Once Toplink DCIM manages the smart PDUs in your racks, the power consumption of each data center is totaled in the resource statistics on the dashboard; if all IT equipment is fed from managed PDUs, that total can serve as the denominator, metered at PDU output. Which rack and rack unit each device occupies is recorded in rack and asset management.

For the first limitation above, the same statistics page shows how many servers are powered on, powered off or in an unknown state. Combine that with servers whose traffic graphs have shown almost no traffic for a long time, and you have a list of machines that are probably idle; confirm each one, then decommission or consolidate. Afterwards PUE may rise slightly while total energy falls, so report both numbers together. For how DCIM and facility monitoring split this data between them, see what is DCIM.