Executive Summary

• The golden rule of no liquids in the server room is a thing of the past, especially as liquid cooling becomes the baseline cooling solution for AI high-density workloads, and now we’re dunking entire racks into fluid with immersion cooling – the irony is not lost on us.
• Immersion cooling does exactly what it says on the tin: your most expensive, high-performance hardware is dunked into a huge tank of non-conductive, dielectric fluid.
• The air vs liquid vs hybrid cooling debate will come down to power effectiveness, and immersion cooling uses less energy than air and in the long-run, makes your data centre more sustainable.

We’ve gone from bringing in your cup of coffee into data centres being a sackable offence, to now fully submerging racks into non-conductive liquids to cool them down.  For sure, the tables have turned and banning liquids in data centres is most certainly a rule of the past, and we can only bask in the irony that is immersion cooling and how it’s a far cry from the no liquids in the server room rule.

What is immersion cooling?

Immersion cooling does exactly what it says on the tin: your most expensive, high-performance hardware and dunks it into a huge tank of non-conductive, dielectric fluid like a bourbon biscuit in your tea.

Say goodbye to fans, screaming air cons and “no liquid” signs because to keep the hottest chips from melting down, the oldest rule has been completely redesigned.

There are two types of immersion cooling: single and double- phase cooling. The term “phase” refers to states of matter and not actual, physical stages in the system.

• Single-phase cooling – the heat from immersed server components is transferred directly to the surrounding fluid. The dielectric fluid doesn’t change state from liquid to gas; the fluid is cycled out of the tank by a coolant pump running through a heat exchanger, which cools the fluid down and returns it to the immersion tank, where it continues the cycle of heat transfer.
• Two-phase cooling – The heat from those immersed server components causes the fluid to boil and change state, from liquid to gas. So the steam heats a condenser coil in the top of the sealed chamber; the coolant in said oil is cycled out of the chamber to a heat rejection mechanism (a cooling tower, for example). Then the coolant is sent back to the sealed chamber at a lower temperature to continue the cycle.

Benefits of liquid immersion cooling

The air vs liquid vs hybrid cooling debate will come down to power effectiveness, and immersion cooling uses less energy than air and in the long run, it’s a recommended method for improving the sustainability of data centres.

Furthermore, while traditional servers with air cooling need air conditioning units, there’s no need to have these huge CRAC units taking up space when you have immersion cooling. The need for hot and cold aisles and raised floors also diminishes. Without those CRAC units and fans whirring, you get a bit of peace and quiet, also – deploying immersion cooling reduces data centre noise.

Liquid immersion cooling can be installed anywhere with power and network hookups and they also increase rack density. A standard 42U cabinet holds 5-7 kW worth of server compute, but fully populated 42U immersion racks can hold over 480kW of compute – this is a rather large selling point for AI data centres with the demand for hardware to cope with high-density workloads.

You’re also future-proofing and helping your cooling strategy evolve to handle complex, dense workloads as we see AI demand predicted to soar still over the next few years. You’ve not got a runway to introduce next-gen products without needing to upgrade as servers evolve.

With data centre cooling strategies, as workloads evolve, so too, must the innovative cooling solutions, and immersion cooling won’t have a hard time keeping up as AI’s power appetite increases.