What is Liquid Cooling for Servers?

What Is Server Water Cooling and How Does It Work?

 

AI servers place multiple high-power CPUs, GPUs, and accelerators within a limited space. As these components generate more heat, increasing fan speed and airflow may no longer provide the cooling capacity required for high-density deployments.

 

Water cooling offers another way to remove this concentrated heat. As one form of server liquid cooling, it uses water or a water-based coolant to collect heat and carry it away from the servers. This article explains how server water cooling works, the equipment it requires, and how it differs from the other liquid-cooling methods.



What Is Server Water Cooling?

 

Server water cooling is a type of liquid cooling that uses water or water-based coolant to collect heat from servers and carry it to another part of the cooling system. Depending on the system design, the coolant may collect heat close to CPUs and GPUs or from the heated air discharged by multiple servers.

 

Water-based coolants generally have a much higher volumetric heat capacity than air. 

For a given temperature increase, the same volume of coolant can therefore absorb considerably more heat. This makes water cooling suitable for removing concentrated heat from high-power servers, where airflow alone may not provide sufficient cooling capacity.

 

Water cooling falls within the broader category of server liquid cooling. Water-based coolants may include DI water or a water-propylene glycol mixture such as PG25. Other liquid-cooling systems may use dielectric fluids.  

 

What Are the Main Types of Server Water Cooling?

 

Server water-cooling systems can collect heat at different points within the server and rack environments. The two main approaches are rear-door heat exchangers and direct-to-chip water cooling.

 

  • Rear-Door Heat Exchangers

 

A rear-door heat exchanger replaces the standard rear door of a server rack with a door containing a water-cooled coil. The servers still use fans to move air across their components. When the heated exhaust air leaves the servers, it passes through this coil.

 

Water or water-based coolant circulating through the coil absorbs heat from the passing air and carries it to the facility’s heat-rejection system. After releasing the heat, the cooled liquid returns to the rear-door coil, and the cycle repeats.

 

  • Direct-to-Chip Water Cooling

 

Direct-to-chip water cooling uses cold plates attached to heat-generating components such as CPUs and GPUs. Water or a water-based coolant flows through channels inside the cold plates and absorbs heat conducted from the chips.

 

The heated coolant leaves the server and carries the collected heat to a coolant distribution unit or heat exchanger. After transferring the heat to the facility’s heat-rejection system, the cooled liquid returns to the cold plates, and the cycle repeats.



What Equipment Does a Direct-to-Chip Water-Cooling System Need?

 

A direct-to-chip water-cooling system uses several connected components to collect heat from processors, circulate the water-based coolant, and transfer the heat toward the facility-side cooling system.

 

  • Cold Plates

Cold plates attach to CPUs, GPUs, and other high-heat components. Water-based coolant flows through internal channels and absorbs heat from the chips.

 

  • Coolant Distribution Unit

The coolant distribution unit, or CDU, controls the temperature, pressure, and flow of coolant supplied to the servers. Its heat exchanger transfers heat from the server-side coolant to the facility-side water without mixing them.

 

  • Pumps, Manifolds and Piping

Pumps circulate the coolant, while manifolds distribute it among multiple servers or racks. Piping and hoses connect these components to the cold plates and CDU.

 

  • Facility-Side Heat-Rejection Equipment

A dry cooler, cooling tower, chiller, or another heat-rejection system releases the collected heat. The equipment used depends on the facility design, cooling load, and local climate.

 

How Does a Direct-to-Chip Water-Cooling System Work?

 

The equipment described above forms two connected circulation loops. The server-side loop, also called the secondary loop, collects heat from the servers. The facility-side loop, or primary loop, carries that heat to the facility’s heat-rejection equipment.

 

The CDU pumps cooled server-side coolant through the manifolds, which distribute it among the servers. Inside each server, the coolant flows through cold plates attached to CPUs and GPUs. Heat moves from the chips into the cold plates and is absorbed by the coolant.

 

The heated coolant then returns through the manifolds to the CDU. Inside the CDU, the server-side coolant and facility-side water flow through separate channels in a heat exchanger. Heat passes from the warmer server-side coolant to the cooler facility-side water, but the two liquids do not mix. After releasing its heat, the cooled server-side coolant flows back to the cold plates, and the server-side cycle begins again.

 

The facility-side water carries the transferred heat from the CDU to a dry cooler, cooling tower, chiller, or another heat-rejection system. After the heat is released outdoors, the cooled facility-side water returns to the CDU, completing the facility-side cycle.

 

How Does CreAlien W2PC Differ from Conventional Direct-to-Chip Cooling?

 

Conventional direct-to-chip systems usually use single-phase cooling. In a single-phase system, the coolant remains liquid and becomes warmer as it absorbs heat. To carry away a higher heat load, the system may need to circulate more coolant or allow the coolant temperature to rise more as it passes through the cold plates.

 

CreAlien W2PC instead uses DI water in a sealed, negative-pressure loop. The negative pressure lowers the water’s boiling point to approximately 45ºC. As the DI water absorbs heat from CPUs and GPUs, part of it changes into vapor.

 

During this phase change, the water absorbs latent heat. The vapor then travels to the condensation side, where it releases the collected heat and returns to liquid. The DI water flows back to the cold plates, and the cycle continues.

 

As the chip heat load rises, more DI water can vaporize to absorb the additional heat. W2PC can therefore support high-heat-flux processors without depending only on a higher coolant flow rate or a greater temperature rise.

 

Contact CreAlien to discuss how W2PC can support your AI server liquid-cooling architecture.