Critical Cooling Infrastructure for Data Centers and Industrial IT Environments

A CRAC (Computer Room Air Conditioner) saját kompresszorral és hűtőközeggel dolgozik, a CRAH (Computer Room Air Handler) viszont nem hűt önállóan: hűtött vizet kap egy központi chillertől, és csak a hőcserét és a levegőmozgatást végzi.
Kisebb szervertermekben a CRAC egyszerűbb és önállóbb. Nagyobb, több gépteremből álló adatközpontban a CRAH + chiller kombináció hatékonyabb, mert a hűtés központosítható és free coolinggal kiegészíthető.

Precision cooling is required in any space where IT equipment operates continuously: it keeps both temperature and humidity within a narrow range year-round, even in freezing outdoor conditions. Comfort air conditioning is designed for human occupancy, typically does not operate in winter, and does not control humidity.

A comfort air conditioner installed in a server room typically fails in two ways: in winter, the outdoor unit shuts down or freezes, and uncontrolled humidity can lead to condensation or electrostatic discharge.

PUE is the ratio between total facility energy consumption and the energy consumption of IT equipment. An older server room with traditional cooling typically operates between 1.8 and 2.2; a well-designed data center supplemented with free cooling is usually around 1.2–1.4; and the best large-scale facilities operate below 1.1.

PUE is largely determined by cooling, which is the second-largest energy consumer after IT equipment. Most savings therefore come not from new servers, but from organizing airflow and improving the cooling strategy.

It is generally worth switching to liquid cooling above a power density of 15–20 kW per rack: beyond this point, air cooling can only maintain temperature with disproportionately high airflow and noise levels. In AI and HPC environments, where a single rack can draw 40–100 kW, liquid cooling is not an option, but a requirement.

You can read more about the structure and operation of liquid cooling systems in our separate article.

Free cooling activates when the outside temperature is low enough for the cooling system to operate partially or fully without compressors. In Hungary’s climate, this can be used for several thousand hours per year, typically resulting in 30–50% savings in the electricity consumption of cooling.

The actual savings depend on the configured water or air temperature: the higher the temperature at which the server room can operate, the more hours the system can run without compressors.

You can read more about the NFrance case study in our separate article.

N+1 means that there is one spare cooling unit in addition to the required units, so the failure or maintenance of one unit does not cause downtime. 2N means a fully parallel system, with independent power supply and distribution. In business-critical environments, N+1 is the minimum; for continuous availability, 2N is required.

You can read more about the relationship between redundancy levels and Tier classification in our separate guide.

Precision cooling should be serviced at least twice a year, and quarterly in high-load or dusty environments. Maintenance should include filter replacement, heat exchanger cleaning, checking refrigerant charge and pressure values, condensate drainage inspection, and review of the control system.

Neglected maintenance does not first appear as a failure, but as increased energy consumption: a clogged filter and dirty heat exchanger reduce efficiency, causing the system to consume more energy for the same cooling output.