Why are so few using immersion cooling?
Are we willing to pay the price?
The data centre industry already has what it takes to achieve a PUE of 1.2 or even ~1.04. But are we ready to pay the price?
Liquid cooling
Liquid cooling has been getting much attention of late, as the world fixates on AI and GPU power needs soar.
Of late, I've been in private briefings and seen various interesting immersion cooling solutions:
- GPU servers in multi-tank full-immersion environment.
- User-accessible, two-phase immersion system.
- Servers 'half-immersed' in dielectric fluid.
- Robot-operated immersion solution.
Incredible efficiency
By taking out noisy, power-hungry fans and leveraging the high heat capacity of liquids, we can eke out maximum energy efficiency for incredible PUEs.
Dunk the servers entirely in liquid, and it gets even better.
But there is a price to pay.
That's because current servers on the market were never designed for an immersion environment. This means custom retrofitting work and proprietary cases or racks.
Price to pay
There are several issues I hear of:
- Cost.
- Warranty.
- Aversion to risk.
- Resistance to change.
- It's just more work.
I'll talk about these another day. But the fact of the matter is: Getting a very low PUE is entirely possible today.
Multiple vendors now offer compelling solutions, but apart from certain hyperscalers, uptake appears stuck in the POC stage or smaller deployments.
Could it be that we are just not ready to pay the price yet?
