Direct-to-chip
Liquid cooling that transfers heat from major components through cold plates connected to a fluid loop.
Also called: Cold plate cooling, DLC, Direct liquid cooling
In practice
Direct-to-chip puts the cooling where the heat is made. A cold plate sits against the processor or accelerator, fluid passes through it, and the heat leaves in liquid rather than being handed to air first. The thermal path is short, which is why the approach scales to densities that room air cannot serve.
The loop that serves the cabinets is usually separated from the facility loop by a coolant distribution unit, which controls the supply temperature and flow the equipment sees and keeps the two fluid systems apart. That unit is a dependency worth drawing: it has power, controls and a failure behaviour, and everything downstream of it shares them.
The part most often underestimated is what is left over. Memory, drives, power conversion and network components are generally not on cold plates, so a meaningful fraction of the cabinet’s heat still enters the air. Deployments that state a capture fraction and design the air side for the remainder behave as expected; deployments that assume liquid handles everything discover a hot room.
Practical questions follow the fluid: connection types and whether they can be made and broken under pressure, leak detection and what it triggers, fluid chemistry and treatment, filtration, and how a technician services a node without draining a row. None of these are exotic, but all of them are new work for an estate that has only ever cooled with air.
Scope of this definition
This is an educational summary of how the term is used in data center practice. It is not a standard, a specification or engineering advice, and where a real decision depends on it, the current adopted standards, verified site information and qualified professional review are the correct sources.