If you’ve ever added engine coolant to an automobile, you’re probably already familiar with the concept of liquid cooling. If not, the idea is straightforward: A liquid coolant, usually water-based, is circulated through the engine block to help prevent overheating. Once that coolant gets hot, it passes through a radiator, which uses external air to help cool the water so that it can be cycled back into the engine to absorb more heat.
The core advantage of liquid cooling over air cooling is that water can absorb heat more efficiently - up to thousands of times more efficiently. Case in point, human beings’ ability to sweat is believed to have been an evolutionary advantage over panting mammals in the hunter-gatherer days, as our ancestors chased prey over great distances and tired them out. Early hominids could expel heat better than their non-perspiring prey, which made them superior long-distance runners.
Now, tens of thousands of years since prehistoric times, the question for data center operators is this: Will liquid cooling provide a competitive advantage as rack-power densities increase?
What is liquid cooling in the data center?
In theory, liquid cooling in the data center is simple: Instead of air, fluids are the primary means for capturing and expelling heat created by the IT load.
Modern processors are smaller and more powerful, which translate to heat load that is more densely packed together in the data center. This higher heat density necessitates more efficient cooling strategies.
Cue liquid cooling. There are two main types and both, in theory, have potential to be far more efficient than traditional cooling methods.
This method is a significant departure from liquid cooling methods of the internal combustion engine. If anything, it’s more comparable to how nuclear rods are cooled. The data center’s IT equipment is literally submerged inside an electrically non-conductive bath—so definitely not water—that can then absorb the heat. The main heat-producing elements (especially the chips) directly discharge the heat to the liquid. There are no server fans, so the IT equipment itself uses less energy. This method is more cutting-edge since it requires changes to server (removal of fans) and data center rack design, not to mention special commercially-available liquids.
Direct to chip
Direct-to-chip liquid cooling is more or less exactly what it sounds like and is much more akin to modern cooling methods in internal combustion engines and computers alike. Small hoses bring cool water to heat sinks and circulate the warmed water to a heat exchanger. These servers typically still have server fans and still require some air cooling to eliminate residual heat generated by electronic components other than the CPU’s.
What is the viability of liquid cooling?
Liquid cooling’s viability depends on the type employed. Immersion cooling is the less popular option. It requires a special non-conductive coolant (usually an oil), so the coolant is more expensive than water. More importantly, it requires significant changes in data center design and operations. On a very basic level, submerged IT loads are only accessible via the top, which means you’ll need specialized servers. What’s more, the coolant can be easily spilled, causing safety hazards for personnel. Finally, in order to replace hardware components, one must pull the server out of the liquid and let it dry sufficiently to remove the cover and replace the component.
Direct to chip, meanwhile, requires soft water, as the minerals from plain old tap water would invariably build up and cause blockages. Water presents a risk when passed through IT equipment and most vendors claim to have means to reduce or prevent water leaks.
Nevertheless, direct to chip is far more pragmatic and therefore often preferred to immersion cooling. Additionally, it’s undeniably more efficient than air cooling, thanks to water’s thermal conductivity.
In this sense, liquid cooling has a place in the future of the data center, but how far into the future depends on several factors:
- Not all servers are currently compatible with liquid cooling; this may mean greater upfront costs.
- Liquid cooling is not nearly as common a technology as air-cooled servers. The familiarity manufacturers and operators have with air-cooled servers make them the default choice.
- Liquid cooling introduces new risks and uncertainties that many organizations are neither ready, nor qualified to make.
- Liquid cooling is not always cost-effective in regions with low-cost energy.
It is absolutely fair to project that liquid cooling is the future of cooling, but that future is more distant than some pundits would have you believe.
As of mid-2019, traditional cooling methods like indirect evaporative cooling—which mirrors perspiration in that it adds water to the air to help lower the temperature—are highly efficient and cost effective, especially in regions with low-cost energy (like Central Washington).
Furthermore, AI and machine learning have the potential to make conventional cooling strategies even more efficient (details on how that works here).
In the meantime, though, Sabey Data Centers will be keeping a close eye on the development of liquid cooling.