“How long will single-phase cooling live?” Dr. Mohammat Tradat, Nvidia’s Manager of Data Center Engineering, asked his audience at Data Center World 2024. “It’ll be phased out very soon…and then the need will be for two-phase, refrigerant-based cooling.” 

This sentiment, this subtle yet profound shift towards two-phase direct-to-chip cooling, has been echoed throughout the industry in recent months. At the same conference, Intel’s Dr. Dev Kulkarni warned data center operators that “If you go all out on single-phase only, you might find you need to switch some infrastructure to two-phase technologies in the same period,” emphasizing his earlier findings on the “demonstrated advantages of the two-phase system” to enable High-Performance Computing (or HPC).

According to Dr. Ali Heydari, Nvidia’s Director of Technical Design, single-phase has 5-7 more years of implementation before the inevitable arrival and popular adoption of two-phase cooling. But what, exactly, makes two-phase cooling so much better than its single-phase counterpart? How can it surpass the limitations of single-phase? And what unique benefits will it unlock for the future of data centers?

The Basics

Two-phase’s advantages become apparent with a closer look at its fundamental mechanics. Unlike single-phase, which relies on a primarily water-based liquid for its cooling, two-phase utilizes a liquid (specifically, a dielectric refrigerant) and, when the liquid boils, its resulting vapor as a coolant. The phase change itself—i.e., liquid boiling and becoming vapor—plays a pivotal role in heat removal.

Utilizing both liquid and vapor is two-phase’s critical source of strength—and single-phase’s biggest limitation. If you rely on water as a coolant only in its liquid form, it’ll gradually warm up until it reaches its heat capacity. At that point, it’ll become too hot to cool anything effectively.

However, if you allow your cooling system’s liquid to boil, its phase change from liquid to gas requires a lot of energy, which means a lot of heat will be absorbed from the nearby surroundings. This continual shift from liquid into vapor will allow substantial heat absorption without a substantial increase in temperature . Furthermore, choosing a coolant with a relatively low boiling point allows fine-tuning the system to operate at the optimal temperature range for processors.

The Benefits

The simple physics of boiling (or nucleation) unlocks a wealth of ripple-effect benefits for data center cooling across a variety of mission critical environments. These benefits include:

Better Heat Removal
Beyond the higher capacity for heat absorption thanks to the phase change process described above, research has indicated that, if all else is held equal (such as wattage, facility water temperature, etc.), two-phase offers greater headroom before the GPU or CPU is “throttled,” which can lead to costly underutilization of expensive processors. Overall, two-phase’s headroom aligns with increasing server and rack lifecycles, addressing today’s needs while future proofing these multi-million-dollar dense rack-level investments, providing confidence that future generations of GPUs can be cooled. In addition, exposure to rapidly varying thermal bands—essentially thermal shock—is all but eliminated, contributing to a longer component lifecycle.

Non-Conductive Refrigerant
Due to each fluid’s boiling temperature, pressure, and vaporization properties, single-phase requires a water mixture and two-phase requires a dielectric refrigerant to accomplish their respective processes. 

While water may be useful as a coolant in other (non-mission critical) environments, it can be hazardous for data centers. That’s mainly because water is conductive, which means it causes serious damage to nearby electronics if leaks occur. Given that single-phase systems require a higher flow rate (more on this later), which increases the chance and severity of leaks, and given that servers cost hundreds of thousands of dollars and racks can cost upwards of millions, relying on water to cool key infrastructure can quickly become a recipe for disaster.

Meanwhile, a two-phase dielectric refrigerant can remove greater heat without the risk of a catastrophic leak. Furthermore, according to Vlad Galabov, Omdia’s Research Director of Cloud and Data Centers, critical heat flux is higher at the boiling point of a refrigerant, which allows processors with greater power to be cooled safely.  Ultimately, safety should be a key concern when choosing your data center’s coolant—which is why Accelsius utilizes a “non-forever,” non-flammable, non-corrosive dielectric refrigerant with an A1 safety rating from ASHRAE and a Global Warming Potential (GWP) of <1.

Lower Flow Rate
Another method that single-phase cooling may use to compete with two-phase’s cooling potential is by increasing the cooling system’s flow rate, or the quantity of fluid that passes through a point in the system in a specific period of time. While it will improve cooling capacity (up to a point), an increase in flow rate is a temporary solution at best; over time, it causes strain on cooling infrastructure, which can lead to higher risk of leaks, corrosion as well as the need for larger, costlier, less energy efficient pumps. 

Compared to single-phase cooling, two-phase cooling uses 1/10th of the flow rate, which mitigates these risks or potential financial burdens. 

Ease of Maintenance
One of water’s final drawbacks, at least compared to a dielectric refrigerant, is its greater need for maintenance. Without frequent observation, the water chemistry and purity of a single-phase system will lead to eventual corrosion and bacterial growth, which can significantly hinder a system’s capacity for heat transfer and negatively affect other key functions. Water requires periodic testing, which often means the collection and third-party testing of water compatibility.  Specialized training is also required to accurately observe a cooling system’s water supply, which may exacerbate your data center’s skills gap. Refrigerants, meanwhile, require less time and resources for proper upkeep.

The Bottom Line

With all these benefits in mind, it’s no wonder that industry leaders have increasingly embraced two-phase direct-to-chip liquid cooling as the long-term solution for data centers. As power demands skyrocket and cooling struggles to keep pace, it’s become imperative for data center operators and managers to seek innovative solutions that minimize hassle, risks, and stress while keeping critical infrastructure cool. When compared to its single-phase counterpart, the research couldn’t be more clear: two-phase liquid cooling isn’t just here to stay—it’s here to succeed.