Skip to content
LIQUID COOLING / NOTE 006

Setpoint control in direct-to-chip liquid cooling.

A direct-to-chip liquid cooling loop has two main control inputs: coolant supply temperature and flow. This note explains what each one does, which limits apply and why the best setting changes with load.

arflo Engineering · · 5 MIN

The loop in brief.

In direct-to-chip cooling, cold plates sit on the hottest components and a liquid carries their heat away. A coolant distribution unit (CDU) separates two circuits: the secondary loop that runs through the racks and the facility loop that rejects the heat to chillers, dry coolers or a heat-reuse system.

Cold plates do not capture all of the heat in a server. Memory, power supplies and other components are often still air cooled, so most liquid-cooled halls run an air system in parallel. Any optimization has to consider both.

Two control inputs, one energy balance.

The heat a liquid stream removes equals mass flow times specific heat capacity times its temperature rise. For a given heat load, less flow means a larger temperature rise across the cold plate, and more flow means a smaller one.

Supply temperature sets the starting point of that rise. Chip temperature then follows from supply temperature, the coolant temperature rise and the thermal resistance between coolant and chip. Supply temperature and flow are therefore two ways of buying the same thermal margin, at different energy costs.

What each setting costs.

Flow costs pump power, which rises steeply with flow rate. Running more flow than the load requires is a common and expensive safety margin.

Supply temperature determines how the heat can be rejected. A warmer loop can use dry coolers or free cooling for more hours of the year and makes the heat more useful for reuse. A colder loop needs more mechanical cooling. The energy-optimal operating point therefore tends toward warm supply and moderate flow, which is also the point with the least thermal margin.

The limits that bound the choice.

Chip temperature limits set the upper bound for supply temperature at a given load. The dew point of the room air sets a lower bound: coolant lines below it collect condensation. Hardware vendors specify permitted supply temperatures, flow rates and pressure differences for their cold plates.

Hydraulics add constraints of their own. Racks on the same loop share a pressure difference, so a change made for one rack affects its neighbours. A setpoint is only valid if every branch still receives its minimum flow.

Why fixed setpoints leave energy unused.

A fixed supply temperature and flow are chosen for peak load. At partial load the same settings produce a margin that is larger than required. Adjusting the setpoints with the load recovers part of that margin as lower pump power and more hours without mechanical cooling.

Doing this safely requires knowing how close each chip is to its limit under the current load, which is a modelling and telemetry question. arflo connects thermal physics with adaptive AI to describe that relationship from chip to infrastructure. What can be saved on a particular loop depends on its hardware, climate and load profile and has to be measured there.