PUE Optimization For Data Centers: Maximizing Compute Efficiency

An airflow and thermal study identifies hotspots and inefficiencies at the room level — typically a snapshot at one or a few operating conditions. A full energy/PUE model layers in energy consumption across cooling equipment and, ideally, a full year of local climate variation, to project annual PUE and, where relevant, WUE and CUE impact.

By identifying specific inefficiencies — over-cooling low-density areas, airflow bypass losses, imbalanced CRAC operation — simulation enables targeted improvements to cooling delivery. Reducing unnecessary cooling load directly improves PUE and lowers long-term operating costs.

It's a starting point, not a strategy. Static math assumes one fixed condition; real facilities run under varying IT load, seasonal climate shifts, and partial failure states. Dynamic, climate-driven modelling captures that variation, which is where most real inefficiency hides.

Where a facility's roadmap includes higher-density AI/GPU racks, we can model the energy case for air cooling versus a liquid cooling retrofit — including hybrid approaches like rear-door heat exchangers or in-row CDUs — as part of the same engagement.

It depends on facility complexity, the number of scenarios required, and how complete the input data is — we'll scope this with you at the start of the engagement. Projects can start immediately on PO, without a separate mobilisation period.

The more complete the data, the more accurate our analysis and optimization. To calculate or model an accurate facility PUE, we will require the following information:

* Electrical & Metering Data:

Single-line diagrams (SLDs) or defined sub-metering points to separate IT load from UPS, transformer, and distribution losses.

*Central Plant Specifications:

Ratings and efficiency curves for primary cooling equipment, including chillers, cooling towers, fluid coolers, and CDUs.

*In-Room Cooling Data:

CRAC/CRAH/FWU unit specs, air/water supply setpoints, and fan power requirements.

IT Rack Configuration:

Floor plans, rack layout drawings, and average or peak IT power load per rack.

*Facility Utilization:

Current total IT load consumption versus total designed facility capacity.

Site Weather Data:

Local hourly weather files (.epw) to evaluate heat rejection performance and free-cooling hours.

Operational Hotspots:

Known thermal problem areas, airflow recirculation zones, or capacity bottlenecks.

Yes — PUE (Power Usage Effectiveness) is used for both live facilities and new builds. The specific applications are different depending on  real-time operational monitoring or predictive dynamic modeling.

1. Live Facilities vs. New Builds
New Builds (Design & Simulation): PUE is calculated using dynamic annual simulations (e.g., IESVE modeling over 8,760 hourly climate points) to right-size mechanical equipment, evaluate free-cooling potential, and ensure regulatory compliance before construction.

Live Facilities (Operations): PUE is tracked continuously via DCIM/BMS metering to measure actual operational efficiency, verify utility bill impact, and meet sustainability standards.

2. How the Specific Use Cases Apply
Troubleshooting & CRAC Rebalancing: In Live Operations: Spikes in real-time pPUE (partial PUE) highlight localized cooling inefficiencies, airflow short-circuiting, or underperforming CRAC/CRAH/FWU fans.

In Simulation/Modeling: CFD and dynamic thermal models simulate hall airflow, velocity, and cold aisle conditions to identify hotspots and rebalance cooling air distribution without making risky physical adjustments in a live hall.

3. Expansion Planning & Failure Scenario Testing

Without Requiring Downtime: Physical testing of failure scenarios (e.g., pulling a chiller offline during a 37.4 °C peak ambient day) on a live, mission-critical facility poses severe risk to uptime.

The Modeling Solution: Engineers use dynamic software simulation tools to stress-test extreme weather return periods (n = 20-50years), test N+1/N+2 equipment failure scenarios, and model hybrid liquid-cooling retrofits completely risk-free prior to live deployment.