Direct-to-Chip Versus Rear-Door Heat Exchangers: How Designers Choose
Two common ways to bring liquid cooling to dense racks, and the questions that separate them.
Key takeaways
- Direct-to-chip cooling pipes liquid to cold plates on the hottest components. Rear-door units cool the air leaving the rack.
- Many facilities combine both, because direct-to-chip does not remove all of the heat.
- Rack density, retrofit limits, water temperature and operational comfort drive the choice.
As computing racks get denser, air alone becomes harder to rely on. Two liquid-assisted methods are common: direct-to-chip cooling and rear-door heat exchangers. They solve related problems in different ways.
Direct-to-chip cooling
Cold plates sit on the hottest parts of a server, usually processors and accelerators. Liquid flows through the plates and carries heat away through small tubes to a manifold at the rack. From there it connects to a coolant distribution unit, often called a CDU, which separates the server loop from the facility water loop and controls temperature and flow.
Because the liquid captures heat at the source, it can handle very high densities. However, it typically does not capture everything. Memory, storage, power supplies and other parts still give off heat to the air, so some air cooling remains.
Rear-door heat exchangers
A rear-door unit is a coil mounted on the back of the rack. Warm air leaving the servers passes through the coil, and water carries the heat away. The room receives air that is close to room temperature, or in some designs cooler.
Rear doors can be added to existing racks without changing the servers, which makes them attractive for retrofits. They have limits as density climbs, and some are fitted with fans to help.
How designers compare them
- Density: direct-to-chip generally scales higher. Rear doors suit moderate to high density.
- Server changes: direct-to-chip needs liquid-ready hardware. Rear doors work with standard servers.
- Water temperature: direct-to-chip can often use warmer water, which helps heat rejection. Rear doors typically need cooler water to be effective.
- Leak risk: direct-to-chip puts fittings near the electronics, so connector quality, leak detection and procedures matter. Rear-door piping sits outside the server.
- Serviceability: hot-swappable quick disconnects help, but staff need training either way.
- Cost and schedule: depends on the number of racks, piping changes and equipment lead times.
Combined approaches
Many designs pair direct-to-chip with a rear door or with air cooling for the remaining heat. The planning question is what share of the rack heat goes to liquid and what share stays in the air.
Water quality is another shared concern. Cold plates have very small channels, so the coolant must be clean and chemically controlled. Filtration, fluid testing and flushing procedures during commissioning protect the equipment and keep flow steady over time.
Questions to ask
- What is the planned rack density now, and what could it become?
- Will servers be liquid-ready at purchase or retrofitted?
- What facility water temperature does each option need?
- How is the liquid loop filtered, treated and monitored?
- What leak detection and shutoff strategy is in place?
- How will the cooling plan adapt if density rises in a later phase?
No single method wins in every case. The right choice depends on the load, the building and the people who will operate it.
This article is general information for public education. It is not legal, tax, financial or engineering advice. Rules and conditions differ by location, so check with qualified local professionals. Spot an error? Write to info@datacenteroutreach.com and we will correct it.