Microsoft Experimenting Throwing Data Center Racks Into Boiling Liquid

In Microsoft’s data centres it could soon become a common sight that server racks are stored in a boiling liquid. It would be the first company to use this method on the scale of a data centre.

Immersing computers completely in a coolant is not a particularly new idea. Such concepts have been used repeatedly in the past, but never to the extent that Microsoft intends to do now. It was more about individual systems or smaller groups of boards. The latter is more common among crypto miners who want to cool a number of graphics cards efficiently.

Read This: Github Actions Would Prevent Hackers To Use Github Projects For Cryptomining

Microsoft is experimenting with dipping complete racks in a fluorocarbon liquid, reports the US magazine The Verge. This is not conducive and comes into direct contact with all components to be cooled. The medium has a boiling temperature of only 50 degrees Celsius, which makes it look a little as if the computers are constantly standing in boiling water. The fluorocarbon vapour condenses again on a cooler outer wall and rains back into the cooling basin so that a closed circuit is obtained here.

Saves Energy Reduces Wear

The process makes it a little more complicated when a component needs to be replaced quickly. In such cases, classic air cooling is easier. However, the tried and tested method offers enormous advantages in other areas. The cooling is much more efficient and energy-saving than with traditional methods. In addition, there is the fact that the server components are no longer in contact with the reactive oxygen in the air. This results in significantly less corrosion and wear and tear on the components of the server.

Microsoft is currently testing the process in a smaller data centre in the USA. The results so far have been very positive, however, and the refrigerated containers, in which rows of racks are immersed, will soon also be found in various large data centres. Here they should be used primarily for very computationally intensive tasks such as training AIs.