Data centre cooling: when air is sufficient and when water must take over

01The question in almost every project

Is cold aisle containment still up to date, or are we planning a model that is being phased out? The question has arisen since racks for AI applications began to give off many times the heat of conventional servers. The answer is nuanced: aisle containment remains the basis – but it is no longer the right solution for every rack.

02The typical fallacy

There is often lengthy discussion about cold or hot aisle. In energy terms, both variants are close to each other. What matters is that the supply airflow follows the actual IT load: a control system maintains a slight differential pressure of a few pascals between the cold and warm zones and continuously adjusts the speed of the computer room air handlers. Because fan power rises roughly with the cube of the speed, even a moderate reduction in airflow saves a lot of electricity.

The differences lie elsewhere. The hot aisle keeps the room cool and offers more thermal reserve in the event of a cooling failure, but requires a return air path via the ceiling – more of a new-build solution. The cold aisle is easier to retrofit into existing rows.

03The principle: power density decides

Air has a low heat capacity. Per unit of volume and kelvin, water absorbs around 3500 times more heat. Above a certain power per rack, the required airflows can therefore no longer be moved sensibly. Three stages then follow:

• Rear-door heat exchanger: a water-cooled coil replaces the rear door of the rack and can be retrofitted. The flow temperature must remain above the dew point of the room air, otherwise condensate forms.
• Direct-to-chip: cold plates remove most of the heat directly from processors and graphics chips; the remainder is still air-cooled. A coolant distribution unit separates the rack circuit from the building network; leak monitoring and redundant pumps are mandatory.
• Immersion: the servers are immersed in a non-conductive liquid. This enables the highest densities, but fundamentally changes maintenance and spare parts logistics.

04Warm water is the goal

The ASHRAE guidelines for liquid cooling classify systems according to the maximum supply temperature of the cooling water, from 17 to over 45 °C. The higher this temperature, the more often dry coolers without a chiller are sufficient – and the more valuable the waste heat becomes for heating purposes.

05What this means for your project

• Determine the power density per rack for today and for the planned expansion – it determines the system, not the other way round.
• Plan zones: air-cooled rows for standard IT and liquid-cooled rows for dense racks are not mutually exclusive.
• Provide cooling water connections, load-bearing capacity and leak protection, even if liquid cooling will only come later.
• Clarify waste heat utilisation early. In the EU, data centres with a rated power of more than 1 MW must use their waste heat, provided this is technically and economically feasible; in Switzerland, individual cantons set their own efficiency requirements.

Properly planned, the waste heat becomes a heat source: a large-scale heat pump can raise it to the temperature level of a district heating network. Anergy networks, which take up waste heat directly, are also suitable as consumers.

Takeaway

It is not the direction of containment that decides, but power density: air for standard IT, water for dense racks – and the water temperature as high as possible so that free cooling and waste heat utilisation work.

SEDC AG · Article from our internal training series “Technik Praxis”, abridged for clients and users.