AI hardware is pushing more heat into liquid cooling loops, and graphene coolants are getting attention as a way to move more of that heat with the equipment a data center already has. The numbers quoted for these fluids are measured in very different ways, though. This post explains what the common measurements mean, what our own tests on ThermaGraph™ show, and what they don't show yet.
Why the coolant matters more as racks get denser
In direct-to-chip and rear-door systems, a liquid loop carries heat from the hardware to a coolant distribution unit (CDU). In single-phase immersion, the servers sit in a dielectric fluid. Either way, how well that fluid picks up heat and gives it off affects how hard the pumps and the heat-rejection plant have to work.
Many technology cooling loops run on 25% propylene glycol (PG25) for its freeze protection and corrosion inhibitors. That protection has a cost: glycol mixes carry heat less well than water. Dow publishes a thermal conductivity of 0.485 W/m·K at 50 °C for DowFrost LC25, and water's is higher. So the choice of carrier fluid matters before any graphene is added.
Thermal conductivity and heat transfer coefficient are not the same number
Thermal conductivity (k) describes how well a fluid conducts heat when it isn't moving. It is measured in a small lab cell and belongs to the fluid alone.
The convective heat transfer coefficient (HTC) describes how much heat moves between a surface, such as a cold plate, and a fluid flowing past it. It depends on the fluid, the flow rate and the geometry, so it has to be measured in a pumped loop. It is closer to what your hardware actually experiences.
A fluid's k can change by a small amount while its HTC in a real loop changes by more or less. Viscosity matters too, because a thicker fluid needs more pump power. That is why we report heat transfer coefficient from pumped-loop tests, and put thermal conductivity, viscosity, method and temperature on the signed lab sheet for the fluid we ship.
What our tests show
ThermaGraph™ is graphene at 0.05–0.20 wt% in deionized water, with no glycol, for indoor closed loops. In our internal pumped-loop lab tests it showed a +6% to +18% higher convective heat transfer coefficient than a glycol baseline. That is lab data, not field data.
Part of any comparison against glycol comes from the carrier itself, because water conducts heat better than a glycol mix. The graphene is in addition to that. It's also why ThermaGraph™ is only for indoor, conditioned loops where freeze protection isn't specified. Its freeze point is 0 °C.
ThermaGraph-D™ is graphene in Drakeol white mineral oil, formulated as a dielectric for single-phase immersion. We don't quote a percentage for it. Its heat transfer is reported against the unfilled dielectric, at the same temperature with the same method, on the signed lab sheet for each lot, along with breakdown voltage and conductivity. The +6% to +18% figure applies only to ThermaGraph™, the water-loop coolant.
What a modeled number means
We also publish 20 to 58 kW saved per 1 MW of data center load. That comes from an engineering model of a 1 MW plant at a PUE of 1.30, from a conservative case to an optimistic one. It is modeled, not measured. A model is useful for deciding whether a pilot is worth running. It can't tell you what your own site will see.
What the lab tests don't show yet
- Field results. We don't have measured results from a production data center to publish yet.
- Long-term behavior in your loop. Suspension over time, filter loading and particle count depend on the system, which is why they are tracked during a pilot.
- Materials compatibility. Corrosion coupons go in with every pilot. ThermaGraph™ isn't a substitute for LC25's inhibitor package until the coupon results come back.
- Freeze protection. ThermaGraph™ has none. Loops that need it should stay on a glycol fluid.
Questions to ask about any coolant performance number
- Is it thermal conductivity or heat transfer coefficient?
- What was the baseline fluid, and at what temperature and flow rate?
- Was it measured in a lab, estimated in a model, or measured in a working facility?
- How much graphene is in the fluid, and what happened to viscosity and pump power?
- How long did the test run, and what happened to filters and particle count?
- For immersion fluids: are breakdown voltage and conductivity reported for each lot?
How to test it in your own system
Every evaluation of ThermaGraph™ starts with a 30-day sample in your own system, and you keep the data. For a closed loop, we baseline ΔT, pump power, thermal conductivity, viscosity and particle count, fill, then repeat the measurements at 7, 14 and 30 days and inspect coupons and filters. For an immersion tank, breakdown voltage and conductivity are added to the baseline, and boards aren't energized until the lot's electricals are on the signed sheet.
See the products, comparison tables, one-pagers and a cooling glossary on our data center coolants page. Graphene Integrations makes graphene in Charlotte, NC and puts it into coolants, coatings, lubricants, concrete and composites.
Request a 30-day sample or call 704-659-3682.
