CRAC vs CRAH: when your server room asks for compressor and when it asks for chilled water
CRAC and CRAH are the two main families of precision cooling units for data centers, and the confusion between them is one of the most common sources of error in sizing server rooms. Although the operational result looks similar (cold air at the rack, hot air extracted), the engineering behind each one is different, and the choice between one and the other defines the sizing of the chilled water plant or the refrigeration system, not just the data center.
This article describes what each technology does, in which contexts each one is more efficient, how to size them according to the data center density and the operational criteria that weigh in the decision. The goal is that the reader finishes with a documented decision between compressor and chilled water, not with a generic preference.
What a CRAC is and what it does
CRAC stands for Computer Room Air Conditioning. It is a self-contained unit that combines all components of the refrigeration cycle in a single cabinet: compressor, condenser, evaporator, and fan. The hot air from the rack passes through the evaporator, is cooled by contact with the circulating refrigerant, and is returned to the cold aisle at the desired temperature (typically between 18 and 27°C per ASHRAE, the American Society of Heating, Refrigerating and Air-Conditioning Engineers).
The main advantage of CRAC is its autonomy: each unit operates independently, does not require an external chilled water plant, and is installed by connecting it to the power feed and to heat rejection (outside air in air-cooled versions, cooling tower or dry cooler in water-cooled versions). This makes it ideal for small and mid-sized data centers, regional sites, and operations where infrastructure simplicity weighs more than operational efficiency. The main disadvantage is efficiency: each CRAC operates its own compressor with individual yields, and the set of units rarely reaches the efficiency of a centralized system.
What a CRAH is and what it does
CRAH stands for Computer Room Air Handler. Unlike the CRAC, the CRAH has no compressor or condenser of its own: it is a heat exchanger that receives chilled water from an external plant (chiller), circulates it through a coil, and returns the water at a higher temperature for the chiller to cool it again. The CRAH unit only contains the fan, the filters, the coil, and the controls.
The main advantage of CRAH is the global system efficiency: the centralized chiller operates with much higher yields than an individual compressor, especially at partial loads, which reduces total electrical consumption. The chilled water plant can also leverage free cooling (free cooling with outside air) in temperate or cold climates, saving between 30% and 70% of annual electrical consumption. The main disadvantage is dependency: if the chiller fails, all connected CRAH units lose cooling capacity. That is why CRAH systems require N+1 or 2N redundancy in the chiller.
Technical comparison between CRAC and CRAH
The following table summarizes the operational differences relevant to correctly sizing each system. The values are typical ranges for mid-sized data centers in operation.
| Characteristic | CRAC | CRAH |
|---|---|---|
| Main components | Compressor, condenser, evaporator, fan (all in one) | Fan, filters, coil (no compressor) |
| Final cooling | refrigerant in closed internal circuit | chilled water from external chiller |
| Typical capacity per unit | 12 to 80 kW thermal | 30 to 200 kW thermal |
| Energy efficiency (kW thermal / kW electrical) | 2.5 to 3.5 (average EER) | 5 to 7 with modern chiller + free cooling |
| Required external infrastructure | none or cooling tower | chiller + chilled water network + tower |
| Typical redundancy | N+1 units | N+1 chillers + water loop |
| Total failure risk | one unit down, only affects its zone | centralized chiller down, affects the entire data center |
| Typical application | small and medium sites, edge data centers | mid-sized and large data centers, high density |
When CRAC makes sense
CRAC makes sense in four concrete scenarios.
The first is the small or mid-sized data center (less than 500 kW of IT load) where the complexity of a chilled water plant is not justified.
The second is the edge data center or regional site where simplicity of deployment and maintenance outweighs efficiency.
The third is the fast project: installing CRAC is faster than installing chillers and a chilled water network, which reduces time-to-market. The fourth is the data center in extreme climate where free cooling is not viable (permanent humid heat): there the chiller efficiency drops, and the autonomy of CRAC weighs more than efficiency.
When CRAH makes sense
CRAH makes sense in four different scenarios.
The first is the mid-sized or large data center (more than 500 kW of IT load) where operational efficiency outweighs initial CAPEX.
The second is the data center in temperate or cold climate where free cooling is viable for a significant part of the year: there the chiller only operates when the outside temperature does not allow passive cooling.
The third is the operation where electrical OPEX (operating expense) outweighs initial CAPEX: CRAH with efficient chiller has lower TCO at 5 years in large operations.
The fourth is integration with district cooling or with waste heat reuse systems: CRAH connects directly, CRAC requires adaptations.
Common mistakes when choosing and sizing
Five mistakes concentrate the majority of problems in data centers that choose wrong between CRAC and CRAH. Recognizing them before buying avoids expensive rework.
- Sizing the CRAC as if it were a CRAH: CRAC and CRAH do not have the same capacity per unit nor the same required redundancy. Calculating the plant with factors from the other system leads to undersizing and hotspots.
- Ignoring chiller redundancy in CRAH systems: a centralized chiller without N+1 redundancy is a single point of failure. If it fails, the entire data center loses cooling within minutes.
- Mixing CRAC and CRAH in the same data center without coordination: the two systems operate at different temperatures and pressures. Mixing them without coordination causes pressure differences that degrade central system performance.
- Underestimating CRAC heat rejection in warm climate: an air-cooled CRAC rejects heat to the outside environment. In warm climates, the condenser loses capacity when the outside temperature exceeds 40°C. Sizing for average climate without considering peaks leads to hotspots in summer.
- Buying low-efficiency CRAC for initial savings: the price difference between a 2.5 EER CRAC and a 3.5 EER one is 10% to 20%. The difference in annual electrical consumption is 30%. The initial savings is lost in 18 months.
Sources
[1] ASHRAE — Technical Resources (thermal management guidance) — https://www.ashrae.org/technical-resources
[2] Wikipedia — Air conditioning (background reference) — https://en.wikipedia.org/wiki/Air_conditioning
[3] TIA-942-C — Telecommunications Infrastructure for Data Centers — https://tiaonline.org/product/tia-942-c/
[4] Uptime Institute — Data center industry resources — https://uptimeinstitute.com/
[5] Wikipedia — HVAC (background reference) — https://en.wikipedia.org/wiki/Heating,_ventilation,_and_air_conditioning
