What to Send Us for a CFD Study

A CFD study is only as good as what it is built from. Every item below that you don’t send becomes one of two things: an assumption we state in the report, or a revision after the first results, when the missing fact turns up. This is the list of what we ask for, by study type, so the first model is the right one.

Send what you have. Nobody has all of it, and a study can start without most of it. The point of the list is to find out early what is missing.

Why it matters

Revision rounds are the main schedule risk in our studies. Of the 18 cooling-failure studies we have delivered for data centers since 2023, six went through four or five report versions. We haven’t sorted every one by cause, and some versions followed a design change on the customer’s side. But a fact that arrives late, such as how long the outage lasts, which unit fails, or what a rack really draws, reopens the model, and every scenario has to be rerun.

A fact that arrives early costs a line in the intake.

Data hall

The room

Floor plan and a section through the room, with the dimensions.
Rack layout, including which positions are empty.
Photos of the room, particularly the containment, the ceiling and under the raised floor. A photo often settles a question a drawing leaves open.

Load

The load of each rack, not a row total. Take the row in the drawing: nine racks at 96 kW in total, with one empty position. Three racks draw 24 kW and six draw 4 kW, so the three hot racks need about 3,800 CFM each at a 20 °F (11 K) air rise, while the row average of about 11 kW suggests about 1,700 CFM. The average hides the racks that matter.
Which rack positions are empty, and whether the empty spaces have blanking panels.

Containment

What is installed: panels, doors, roof, end-of-row doors, curtains.
Where the gaps are: door seams, cable cutouts, gaps above the racks. Leakage is a large part of how contained aisles really behave, and drawings seldom show it.

Supply

Raised floor height, and whether the plenum is clear or crossed by cables and pipes.
Tile types, their open area and where they are.
If the room supplies air some other way, such as ducts, wall grilles or overhead, how.

Return

Ceiling height and plenum depth, if there is a plenum.
Return grilles: type, count, size and rated CFM. Check each against its rating, because an overloaded return can look like a cooling shortage.

Cooling units

Make and model, nameplate airflow and capacity.
Supply setpoint, and how the unit is controlled: supply temperature, return temperature or fixed fan speed.
Where each unit is, which way it blows, and what it draws from.

Site conditions

Site altitude. At about 6,000 ft (1,830 m) the air is about 20% thinner than at sea level, so the same CFM carries about 20% less heat. We model the real pressure, but we have to know it.
Design ambient temperature, if the units reject heat outdoors.

The questions

Which failures to model. If you are N+1, tell us which unit matters to you. Each case adds a solve.
Any earlier study, or measured temperatures. Measurements are the best check we can make on a model.

Site and external flow

Studies of a campus or an equipment yard in the wind ask for:

A site plan with elevations and building heights.
The layout of the equipment yard.
For each unit, make, model, airflow, discharge direction and discharge height.
Screens, walls and louvers, with their sizes.
The design ambient temperature, and the nearest weather station for the wind.
Which intakes matter most, and the entering-air limit of each unit if the data sheet gives one.

The heat each unit rejects deserves a check against the data sheet before the study starts. A condenser rejects its cooling capacity plus the compressor work, which is the number the air picks up. A generator’s radiator rejects far more than the engine’s radiated heat, and the smaller number is the one that tends to get entered. Heat out is airflow times air rise, so you can test any entry: if the flow and rise in the model don’t multiply to the data sheet’s heat, one of them is wrong.

Exhaust dispersion

Stack locations, heights and diameters.
Exhaust flow and temperature for each source.
Fuel and emission rates, from the manufacturer’s data sheet.
The intakes to protect, with the airflow of each.
The buildings around the stacks, and the nearest weather station.

The exhaust dispersion service lists what we deliver.

Cooling failure

A cooling-failure study follows the room through the event, so the sequence matters as much as the room.

What is on UPS: the IT load, and whether cooling fans and pumps are.
Chiller restart time, and how long the chillers take to reach full output.
Generator start time.
Thermal storage, if any, and how much.
The event you want to model. Total power loss, a fan-only ride-through and the loss of one unit are different questions.

If you have a CoolSim model of the room that solves in steady state, send it. If you don’t, the data hall list above is what we build it from.

Noise

We don’t run noise studies. If noise comes up on a site we are already modeling, the site editor can give a screening estimate that compares layouts, and for that we need:

The A-weighted sound power of every source, from the data sheet.
The receptor locations: the property line, the nearest neighbors.
The local limit, with the time of day it applies to.

A screening estimate is not something to hand a reviewer. A noise study for a permit needs an acoustician and the manufacturer’s octave-band data for every source; our note on equipment yard noise explains why.

File formats

DWG and PDF drawings are what we work from today. If your drawings are in something else, ask before you export; we would rather tell you what works than get a file we can’t read. A site that exists only as a 3D model is worth a short conversation first.

Vector PDFs and DWG files with the layers intact are much better than scans. If a drawing is a scan, a note of the key dimensions saves a good deal of guessing.

What happens when something is missing

We don’t stop for it. We write down what we assumed, give the source, and say whether the answer would change if the assumption were wrong. On the two drawing-based hall builds we’ve done so far, the open points were ordinary ones: ceiling and floor heights, the cooling unit model and airflow, whether the unit controls supply or return temperature, power labels that disagreed with the notes, and rack orientation. Each goes to you as a short question with our assumption beside it, marked if the answer moves the result.

You can answer them up front, or let us start with the assumptions and send corrections. The first costs a few minutes now. The second costs a rerun.

Next step

Send what you have, even partly, through our contact page, along with what you need to decide. We’ll tell you what’s missing and agree the cases with you before anything is run. The studies page describes the kinds of study we run.