Why external CFDs matter in data centre design

Artificial intelligence is changing the modern data centre, but one thing remains clear: power density is increasing rapidly. Higher rack loads, larger halls, and energy-intensive workloads necessitate cooling systems that can scale quickly and efficiently. Air-cooled chillers remain a popular solution; however, fitting more of them onto constrained rooftop spaces introduces complex challenges. Poor airflow, hot air recirculation, and wind effects can compromise performance in ways that traditional design approaches often overlook. This is where external computational fluid dynamics (CFD) comes into play. 

What is computational fluid dynamics? 

Computational fluid dynamics is a simulation method that solves equations about fluid flow, energy transfer and/or momentum. For rooftop mechanical design, external CFD provides a clear view of how factors like chillers, wind, and heat interact in real-world contexts. Unlike internal CFD models, which look at airflow inside data halls, external CFD focuses on outside factors. This is important because understanding how warm air from equipment like generators interacts with wind flow and affects nearby building temperatures and pressure patterns. 

Engineers model CFD using software based on Navier-Stokes equations. These are a set of differential equations that describe the movement of fluids and air. These insights into the effects of wind uplift of thermal plumes can't be gained from drawings or standard modelling. When AI-ready facilities are being examined, the information from models can significantly affect the cooling performance. 

Why does CFD matter more now? 

CFD has been around for some time, but the growth of AI has made it much more necessary. AI workloads increase rack loads to 25kW or more, sometimes peaking at around 75kW per rack. This means more chillers are needed in buildings and as space is a limiting factor, they are often placed closer together. Combining this fact with the building of more data centres in densely populated cities, and the risk of warm air recirculation increases. 

The problem with recirculated warm air is that it raises the internal temperature of the data centre which reduces cooling capability. Data centres need to be operational at all times, so this can be a key issue as overheating racks can cause them to shut down. Without CFD predicting airflows, it may not be apparent that there is an error in design until the data centre is operational - at which point it is too late. 

How is external CFD applied? 

A CFD study begins with the weather data specific to the site. Designers use ASHRAE's Weather Data Viewer to review variables including dry bulb temperatures, wind speeds, and wind directions. This develops a likely airflow model. 

This model then incorporates the rooftop equipment like chillers, generators, air handling units, and louvres as well as the impact of adjacent buildings. The goal is to simulate real-world implications across different scenarios from best case to worst. Once a preferred option is selected, simulations are assessed under different conditions to test resilience. 

What do the simulations show? 

Once the model has been created, the simulations can take place. Designers will analyse two key results: 

1. Temperature uplift: The increase in intake temperature resulting from recirculated hot air or poor airflow.
2. Recirculation percentage: The proportion of discharged warm air that re-enters the equipment intake. 

These figures are reviewed for each chiller, per data hall, and across the full site. If the results show that there is a high uplift or impactful recirculation, the teams know to edit the designs. Another step involves comparing the simulations against data from equipment manufacturers to find out how chillers are affected. For example, a chiller rated at 1.2MW at 35℃ may face heavy derating at higher intake temperatures. If this capacity could compromise resilience targets or exceed limits, mitigation is required. 

How can you mitigate thermal risk through design? 

The importance of CFD modelling is that it doesn't just diagnose an issue, it provides the information required to fix it. Simulations can guide the solutions. 

In our experience, some strategies involve: 

1. Increasing the space between chillers to improve airflow separation.
2. Raising discharge heights so hot air doesn't pool near intake areas.
3. Installing wind barriers to manage the direction of airflow.
4. Reorienting nearby equipment like generators.
5. Lift generator flues to avoid cross-contamination of heat. 

Lessons learned for the industry

My original paper on this topic found in Ecolibrium Magazine specifically focuses on external CFD in relation to air-cooled chillers. However, the wider impact of external CFD modelling goes further than this. 

Take evaporative cooling systems - they face similar airflow challenges but are sensitive to moisture levels. Recirculating humid air can increase the wet bulb temperature around cooling towers which increases the capacity loss of air-cooled units. This is important as we move towards net zero carbon goals as the impact of water usage in data centre systems will come under further scrutiny. 

External CFD modelling has moved beyond a tool in the kit. Instead, it should be considered a fundamental part of the design process for data centres.