Direct-to-chip, single-phase and dual-phase cooling explained
In the datacenter world, one of the biggest challenges – and one of the greatest opportunities – is heat. When server components operate, all the electrical energy they consume is ultimately converted into heat, which must then be removed. Traditionally, organizations have relied on air cooling, typically using fans to manage this heat. In fact, according to sources, 80% of datacenters still primarily use air cooling to get rid of heat from server components.
However, air cooling is both inefficient and energy-intensive. According to a recent McKinsey study, air cooling can account for 40% of all energy consumption in the datacenter – it’s no surprise, then, that 40% of datacenter operators are exploring alternative cooling methods in search of eco-friendly, cost effective options to support sustainable datacenter operations.
At a fundamental level, this approach makes sense. As anyone who has flown a kite knows, air can be tricky to direct and manage. Fluid is much easier to control– as anyone who’s squirted a water pistol (or been on the receiving end of one) will also know.
However, fluid cooling techniques are changing, especially as we encounter server loads that tend to be more energy-intensive, and as a result, run hotter.
Research and Development Engineer, OVHcloud.
What is direct-to-chip (DTC) cooling?
DTC is the most common form of water cooling and has been used by datacenters and gamers for decades. A metal plate sits on top of the processors (CPU or GPU) with a conducting material between the two – usually a thermal Interface Material (TIM). The plate then uses liquid in pipes to move the heat away from the chip, and a dissipation mechanism to disperse the heat.
This dissipation mechanism can be something as simple as a single fan, although in industrial settings, you tend to see dry coolers equipped with evaporative cooling mechanisms. Dry coolers are units that sit outside the datacenter. They feature radiator-like, finned heat exchangers through which the heated liquid is circulated. Fans on the outside of the units draw cool air in and pass it around the fins, cooling the liquid. This liquid is then fed back into the system and the process is repeated.
In warmer months – or locations – evaporative cooling is used alongside dry coolers. Hot air is drawn through wet pads, evaporating water and cooling the air. This air is then used to cool the liquid from the datacenter through the dry coolers.
These dissipation methods can be used across most types of liquid cooling. DTC is more focused than air cooling, largely because liquid in pipes is easier to direct to specific components than air. With air cooling, although you can direct air currents via fan orientation, DTC allows you to be very precise. It’s also more efficient, thanks to physics, because liquids typically possess higher thermal conductivity than gases.
But even with DTC, some air cooling is usually necessary because of the challenges in adapting multiple cold plate designs to accommodate all the IT equipment that generates heat. GPUs and CPUs do generate the majority of the heat in servers, but RAM and hard drives also get hot, so some air cooling is often used. The ratio of air- to water-cooling is usually in the region of 30%/70%.
However, the world of cooling is constantly moving, and there is another form of cooling which can cool all the components at once.
Immersion cooling
In immersion cooling, the entire server is immersed in fluid. There are several benefits to this: all the components can be cooled at once, and higher heat loads can be handled. Furthermore, because the entire server is immersed in liquid, dust cannot enter the system, substantially enhancing product lifespan. However, immersion cooling is a lot more complex than DTC and maintenance is a more involved process.
There are two forms of immersion cooling – single phase and dual phase. In single phase cooling, the liquid stays liquid throughout the cycle (i.e. it keeps its phase constant). In dual phase cooling, it does not.
Single phase immersion cooling
In single-phase immersion cooling, a cool fluid enters at the base of the immersion unit to cool the server, while the heated fluid leaves at the top – and as with DTC, a dry cooler is used to cool this fluid after it passes through a plate heat exchanger. A separate coolant in a loop is then used to dissipate the heat.
Dual phase immersion cooling
In a dual phase cooling system, servers are immersed in fluid, but the fluid has a low boiling point. When server components heat up, the liquid boils and is directed to a condenser unit, where the gas (vapor) is cooled and re-condensed into a liquid. The fluid can then flow back down into the system to be re-used.
However, not only are the coolants in dual-phase systems generally more expensive, but maintenance is even more difficult, partially because the liquid boils into steam, which is much harder to manage than the liquid in a single-phase system.
Waste Heat
There’s also another significant part of the equation: although we can use dry coolers and evaporative cooling to get rid of waste heat, isn’t there something better we could do with it?
This is a big challenge for datacenters, particularly ones that have been around for a while. A lot of datacenters are on industrial parks, well away from areas where heat can be easily re-used; we’ve all seen the stories about swimming pools being heated using datacenters heat, but this isn’t always practical. Water must be physically transported via pipes to other locations, and then heat exchangers used to heat the other water which results in energy losses, because no energy exchange process is completely efficient.
Additionally, some datacenters face limitations that hinder their ability to reuse waste heat. For instance, in some of our datacenters , the heated water only reaches around 45 degrees; you can safely put your hand on the ‘hot’ pipe. But this also means that the resulting heat is less useful – it’d take a considerable amount of time to heat a swimming pool with a 45 degrees heat source.
However, it’s important to keep pushing forward. In our German datacenters , for example, we don’t use gas boilers to heat our office, instead, we utilize the waste heat from the datacenters because it’s nearby. As an industry, we must continue to advance this innovation. When new datacenter sites are built, organizations should consider heat reuse from the outset.
Over the past two decades, liquid cooling technology has made significant strides and is now capable of managing increasingly high power and heat loads in both personal and industrial settings. Although we are still in the early stages of immersion technology’s evolution, it holds great promise for addressing components that operate at very high temperatures. However, like all technology, it’s not a one-size-fits-all solution, and we’ll almost certainly continue to see a mix of DTC and immersion cooling – not to mention a little air – across datacenter estates for years to come.
We’ve listed the best bare metal hosting.
This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
In the datacenter world, one of the biggest challenges – and one of the greatest opportunities – is heat. When server components operate, all the electrical energy they consume is ultimately converted into heat, which must then be removed. Traditionally, organizations have relied on air cooling, typically using fans to…
