With the liquid cooling market on track to experience tremendous growth, all eyes are on the various types of liquid cooling technologies available today. And a key consideration that data centers have to weigh when deciding which one to use is whether or not they use water. Not only is water a scarce resource that should be highly conserved for people, but it also poses significant risks to hyperscalers because one simple leak can ruin servers that cost $200-300K each and lead to significant unplanned downtime.
In this blog, we’ll take a look at the two main direct-to-chip liquid cooling technologies today known as single phase and two-phase direct-to-chip. While both processes involve the use of cold plates that are placed directly on top of the CPUs and GPUs, the big difference is that single phase uses water in the cold plate while the two-phase approach uses a heat transfer (waterless) fluid.
The Hotter the Chips, The More Water Required
With the single-phase, direct-to-chip liquid cooling approach, water is used as the coolant in the cold plate and it always remains in a liquid state. The ability to take away heat is dependent on water flow, which means that the hotter the processors get, the more water is required to cool them. With next-generation chipsets such as NVIDIA’s GB200 consuming a staggering 2,800 watts of power, it is easy to see that the water flow required to cool chips in the data center in the future will be impractical and against the core sustainability goals set forth for by the entire semiconductor industry.
In contrast, the two-phase direct-to-chip approach used in ZutaCore’s HyperCool technology leverages heat transfer fluid in the cold plate that is 100% safe to the IT equipment and will not damage the server even in the unlikely event of a leak. It also never needs to be replaced, which makes it very easy to maintain over time. In fact, a recent article from Analytics India Magazine (AIM) states that “ZutaCore’s innovative HyperCool technology reduces water usage by up to 90% and energy consumption by 40% in data centers, making it an ideal solution for NeevCloud’s mission to drive sustainable AI adoption.” NeevCloud is one of the latest companies to unlock the power of two-phase direct-to chip liquid cooling and its aggressive plans include the construction an AI cloud infrastructure tailored for Indian clients, comprising 40,000 graphics processing units (GPUs) by 2026.
How Liquid Cooling Works without Using Water
With two-phase, direct-to-chip liquid cooling, the heat from the GPUs and CPUs boils the heat transfer fluid at a low temperature, absorbing the heat using the very efficient phase change physical phenomena keeping the chip at a constant temperature. The process of changing the state of the dielectric fluid from liquid to gas and then back to liquid again is done in a completely closed loop system. As the liquid inside the cold plate boils, the liquid in the cold plate never passes the boiling temp even if the heat increases by 3X (aka hotter GPUs and CPUs). This makes this technique highly scalable for cooling higher power chips in the future
The Risk of Water Leakage in Data Centers
As you can imagine, the use of water in a data center can pose serious risks that can be very costly. Not only can a water leak completely destroy $300K servers, but it can also lead to corrosion, mold, residue, biological growth and other environmental impacts. As the price of server equipment continues to skyrocket with each new technology advancement, it is also feasible for the industry to reach a point where insurance companies either refuse to insure such risk or do so for a crazy amount of money. After all, why use water if you don’t have to?
Another downside of single-phase direct-to-chip being dependent on water flow is that it requires an added investment of larger pipes, tubs and connectors that can resist leakage of water as flow and pressure increases. It also requires larger, power-hungry pumps to continually carry the water through the system. None of this infrastructure investment is needed with direct-to-chip liquid cooling because just a small amount of heat transfer fluid is used, and even if it did leak, it would not harm any equipment. Water is not so forgiving and this was never more evident than in a recent Tweaktown article that reported how “NVIDIA was on the eve of unleashing its next-generation GB200 Superchip inside of new AI servers, until rumors of a "flooding" occurred.”
Sustainability in the Data Center Will Always Win
The ramp up of AI – and the infrastructure to support it – was faster than many people every imagined. This resulted in many data centers and hyperscalers forgoing their focus on sustainability just so they could meet the demand for more processing power. This is all changing now as its becoming clear that sustainability and AI can co-exist by leveraging technologies such as HyperCool that were built with sustainability in mind. Together, we can save the planet and bring generative AI and HPC computing to the masses!
To learn more about HyperCool and how it drives the future of AI sustainability, download this eBook.