Size a data center cooling fan from the IT load and the design air temperature difference, not from floor area: at a 12 K rise you need roughly 2,500 m³/h of air per 10 kW of IT load, and at 15 K about 2,000 m³/h per 10 kW. Then choose EC fans arranged as a redundant array rather than one large fan, because EC motors hold 85–92 % efficiency down to 30 % speed and an N+1 fan wall keeps cooling available during a single fan failure.
Step 1: Convert IT load into airflow
The governing relationship is Q = 3,600 × P / (ρ × cp × ΔT). With air at about 1.2 kg/m³ and 1.005 kJ/kg·K this simplifies to roughly Q ≈ 3,000 × kW / ΔT in m³/h. A 500 kW hall at 12 K therefore needs about 125,000 m³/h, while the same hall at 15 K needs about 100,000 m³/h. Widening ΔT is the cheapest way to cut fan energy, which is why containment and blanking panels are worth more than a larger fan.
ASHRAE class A1 recommends supply air of 18–27 °C with an allowable range of 15–32 °C. Most modern halls are designed at 22–25 °C supply with 35–40 °C return, giving the 12–15 K difference above. Add 10–15 % airflow margin for bypass and leakage even in a contained hall; in an uncontained raised-floor hall the practical margin is 25–40 %, which is itself an argument for containment.
Step 2: Establish the real static pressure
Data center fans are low-to-medium pressure machines, but the numbers vary widely by architecture. A direct-blow computer room air handler with a chilled water coil and a filter typically sees 250–500 Pa. Add an underfloor plenum and perforated tiles and it rises to 400–700 Pa. An air-handling unit serving a hall through overhead ducts with an indirect evaporative heat exchanger can reach 600–900 Pa. Coil face velocity should be kept to 2.0–2.8 m/s: above 3.0 m/s pressure drop rises sharply and noise increases by 3–5 dB.
Step 3: Choose the motor and drive technology
| Option | Efficiency at 100 % / 50 % speed | Turndown | Redundancy behaviour | Best use |
|---|---|---|---|---|
| EC backward-curved plenum fan | 88–92 % / 80–86 % | 10–100 % | Excellent in arrays; failed unit isolates itself | New CRAH, fan walls, retrofit of legacy units |
| IE3 induction motor with VFD | 88–93 % / 72–80 % | 25–100 % | Good, but drive is a single point of failure | Large AHUs above roughly 15 kW per fan |
| Fixed-speed AC belt-driven fan | 80–85 % / not applicable | None | Poor — all-or-nothing | Legacy installations only |
| Direct-expansion in-row unit with EC fans | 85–90 % / 78–84 % | 15–100 % | Very good, granular per-rack response | High-density rows above 15 kW per rack |
The cube law dominates the economics. Running an array at 70 % speed uses about 34 % of full-speed power, and at 50 % speed about 13 %. Because a hall typically operates at 40–70 % of design IT load for the first several years, a variable-speed array often cuts annual fan energy by 50–70 % against fixed-speed operation, and fan energy alone is commonly 5–12 % of total facility power. Fixed-speed AC fans therefore have no place in a new data center, because most halls run well below design load for years.
Redundancy: what N+1 actually means for fans
A single large fan cannot be made redundant; an array can. If a hall needs 100,000 m³/h and you install ten EC plenum fans each rated 11,000 m³/h, you have N+1 at the fan level and the remaining nine simply ramp up when one fails. The speed increase required is modest — going from 90 % to 100 % of array output raises each fan speed by about 11 % and its power by about 37 %, which is why the motors and cabling must be sized for the failure case, not the normal case.
- N+1 at fan level: one spare fan per array. Suitable for most enterprise halls and Uptime Tier II equivalents.
- N+1 at unit level: one spare air handler per group of units. Needed where a whole coil or drive failure must be tolerated.
- 2N: fully duplicated cooling paths with independent power and pipework. Used for Tier IV style requirements and financial trading halls.
- Backdraft dampers on every fan in an array so a stopped fan does not become a recirculation path; without them a single failure can cost 8–15 % of array output.
- Dual power feeds per fan group, with alternate fans on alternate feeds, so losing one feed halves rather than eliminates airflow.
When to choose which fan architecture
- Choose EC plenum fan arrays when the hall is new, when rack density is 5–15 kW and reasonably uniform, and when the operator wants granular speed control tied to cold-aisle temperature sensors.
- Choose in-row or rear-door units with EC fans when density exceeds roughly 15–20 kW per rack, because room-level airflow cannot deliver that locally without excessive velocity.
- Choose a large centrifugal fan with a VFD when a single central air handler serves the hall through ductwork at 700–900 Pa and physical space for an array is unavailable.
- Choose an axial fan only for outdoor heat rejection, free-cooling coils and exhaust duties below about 400 Pa — not for the raised-floor supply path.
- Retrofit existing belt-driven units with EC plug-fan arrays when the coils and cabinets are still sound; this typically returns 30–50 % fan energy savings and removes belt maintenance entirely.
Reliability, noise and maintenance figures
Specify bearings with an L10 life of at least 40,000 hours, and preferably 80,000 hours, because data center fans run continuously — 8,760 hours per year means a 40,000 hour bearing is a 4.5 year component. EC motor electronics are typically the life-limiting item, with published lifetimes of 40,000–60,000 hours at 40 °C ambient and considerably less if installed in a hot return-air stream, so keep the drive electronics on the cool side of the coil where the design allows. Sound levels of 70–80 dB(A) at 1 m are normal for a fan wall at full speed, dropping 8–12 dB at half speed. Keruite manufactures these fan assemblies at an 80,000 m² facility in Dezhou, China, with an in-house AMCA-210-style test laboratory, and every impeller is balanced to G2.5 and performance-tested before shipment.
- Monitor each fan individually over Modbus so a single failure raises an alarm instead of hiding behind array compensation.
- Trend fan speed against IT load; a rising speed at constant load indicates coil fouling or filter loading.
- Keep at least one spare fan module on site, since array capacity is only redundant until the first failure is replaced.
FAQ
How much airflow does a data center need per kilowatt of IT load?
At a 12 K air temperature difference, allow roughly 250 cubic metres per hour per kilowatt, so about 2,500 for every 10 kW. At 15 K it falls to about 200 per kilowatt. Add 10 to 15 percent margin in contained halls and considerably more without containment.
Why are EC fans preferred over AC fans with variable frequency drives?
EC motors integrate the drive, hold 80 to 86 percent efficiency even at half speed, turn down to about 10 percent, and fail individually rather than taking a whole array offline. Below roughly 15 kW per fan they are also cheaper installed than a separate motor and drive.
What does N+1 redundancy mean for a data center fan array?
It means one spare fan beyond the number needed at design load. If ten fans meet the duty, eleven are installed, and the survivors ramp up when one stops. Motors, cabling and backdraft dampers must all be sized for that elevated failure-case speed.
What supply air temperature should a data center run?
ASHRAE class A1 recommends 18 to 27 degrees Celsius supply, with an allowable range of 15 to 32 degrees. Most modern halls are designed near 22 to 25 degrees supply against 35 to 40 degrees return, which gives the 12 to 15 kelvin difference used for airflow sizing.
How much energy do data center fans actually use?
Fan power is commonly 5 to 12 percent of total facility electrical load. Because power varies with the cube of speed, moving an array from fixed speed to variable speed at typical part-load operation cuts annual fan energy by roughly 50 to 70 percent.