Use PUE within its facility and accounting boundaries
Allocate shared overhead transparently without treating an average ratio as marginal cooling savings.
On this page
The question for today
What you’ll understand
- Compute PUE from matched facility and IT energy accounts.
- Apply and reconcile an explicitly chosen proportional allocation.
- Explain why lower PUE does not alone imply lower total energy or marginal job impact.
Electricity delivered to a server is only part of a facility account. Cooling, external power conversion and distribution, and other supporting equipment may consume additional electricity. Power usage effectiveness summarizes a facility-to-IT ratio, but the ratio needs a measurement period and consistent equipment classification. It is useful for accounting and facility analysis, not a universal multiplier describing what would happen if one job disappeared. Keep the metered totals visible so an attractive ratio cannot hide a growing absolute footprint.
Define the numerator and denominator together
The Green Grid’s comprehensive PUE methodology distinguishes total dedicated data-center facility energy from IT-equipment energy. The IT account includes computing, storage, and networking. Facility totals also include the relevant supporting electrical and cooling systems. Match the period, energy sources, and boundaries before dividing. A ratio combining a yearly facility bill with a one-hour IT sample can confound seasons and operating states instead of describing the intended period.
Classification determines where overhead appears. A server’s internal fan can already be inside its IT electrical measurement, whereas a facility cooling system sits outside that IT boundary. Adding both a measured inclusive server value and another estimate for its internal fan double-counts consumption. A shared building may require a documented way to isolate dedicated facility energy. State meter placement and allocation choices rather than inferring them from a label such as cooling power.
A proportional allocation is a declared policy
If a job’s IT energy is known, multiplying by the period’s average PUE is one proportional way to allocate facility electricity. It assigns the job the same fraction of supporting energy as its fraction of IT energy. When every job and any residual IT usage are included exactly once, the allocations reconcile to the facility total. That reconciliation is an accounting property; it does not establish that cooling responds linearly to each job’s instantaneous load.
A facility may have fixed support loads, weather-dependent cooling, and nonlinear equipment efficiencies. Removing a small job could leave much of that consumption unchanged, while a large capacity change could alter the operating regime. Average PUE therefore need not equal marginal facility energy divided by marginal IT energy. Evaluating an intervention needs a response model or measurement at the relevant scale, with conditions stated, rather than automatically interpreting an allocated overhead number as avoidable electricity.
Interpret the ratio alongside absolute service
A lower PUE can coexist with higher total facility energy when the IT denominator grows. That is not a contradiction: the ratio asks how much facility energy accompanies each IT energy unit, not how many useful tasks the site completed. Likewise, replacing efficient IT with wasteful IT can change the denominator in ways that make a facility ratio appear favorable. Compare useful service, IT consumption, and facility totals together when evaluating an infrastructure decision.
For workload comparisons, disclose whether PUE is annual, seasonal, or matched to the measurement window, and whether it is observed or assumed. A fictional scalar is adequate for teaching an allocation calculation, but real claims need appropriate facility evidence. Preserve uncertainty rather than presenting a coarse facility factor as a precise per-request measurement. Finally, check whether an incoming energy estimate is already facility-inclusive before applying any overhead factor a second time.
Work it through
Allocate one facility account without inventing marginal savings
A fictional facility records 100 kWh of IT energy and 130 kWh total facility energy for a period. One job accounts for 8 kWh of the IT total. Apply proportional allocation. Separately, another fictional period has 150 kWh IT and PUE 1.1; compare absolute totals without assuming equal useful work.
The first period has PUE=130/100=1.3 and non-IT energy 130−100=30 kWh. Its numerator already includes its IT energy, so adding 100 kWh to 130 kWh would be incorrect.
The job’s IT fraction is 8/100=0.08. Allocate 0.08×130=10.4 kWh of facility energy, consisting of 8 kWh IT and 2.4 kWh allocated supporting energy. This does not prove that canceling the job saves 10.4 kWh.
The second period totals 150×1.1=165 kWh. Its PUE is lower, yet its absolute facility energy exceeds 130 kWh. Without matched useful work and conditions, the ratios alone cannot rank overall efficiency.
Proportional allocation assigns the job 10.4 kWh. Lower PUE and lower total energy are separate claims, and allocated overhead is not automatically marginal savings.
The living notebook
Explore the model
Interactive lab · matrix
Inspect a proportional scale without a thermal claim
Reset, set Diagonal scale a to 1.3 and Shear b to 0; predict determinant 1.3 before Step. Reset, set a to 1 and b to 0; predict determinant 1 before Step. The selected matrix scales unit-square area by that factor; readouts describe the final matrix while animation interpolates geometry. This is an analogy for a multiplier, not a cooling model or PUE measurement.
- Area scale · det A
- 1.4
- Singular values · σmax, σmin
- 1.597, 0.876
- Condition number · κ₂(A)
- 1.823
The selected matrix is A = [[a,b],[0,1]]. Playback interpolates geometry between the identity and A; readouts always describe A. Its determinant is a and its singular values come from AᵀA. Positive a makes every supported matrix invertible. This two-dimensional family is not a general eigensolver or a learned network. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forA scalar allocation changes totals mechanically but does not explain a facility’s response.
Pause & reason
Check your understanding
Your turn
Try it yourself
For one fictional matched period, facility energy is 60 kWh and IT energy is 50 kWh. A job contributes 5 kWh of disjoint IT energy. Find PUE, allocated job facility energy, and allocated supporting energy.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
Use the job’s fraction of IT energy and keep the inclusive facility total separate from overhead.
Compare with the solution
PUE is 60/50=1.2. The job’s IT share is 5/50=0.1, so proportional facility allocation is 0.1×60=6 kWh. Supporting allocation is 6−5=1 kWh. The remaining 54 kWh belongs to other IT and support allocations. Nothing in these totals alone identifies the marginal savings from removing this job.
Keep the idea
Take it with you
- PUE compares compatible facility and IT accounts over one period.
- Average allocation and marginal facility response are different quantities.
- Report useful service and absolute energy alongside the ratio.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.