What is dielectric fluid? All about server cooling fluids
Key takeaways
- Dielectric fluid is electrically insulating, so it can come into direct contact with powered electronics while carrying heat away from the components.
- Not all liquid cooling uses dielectric fluid. Direct-to-chip systems typically use water or water-glycol inside cold plates, while dielectric cooling uses electrically insulating fluid around the components.
- Precision liquid cooling keeps the fluid contained inside the server chassis, reducing fluid volume and allowing most components to be serviced without draining the coolant first.
Tell someone that a server is being cooled with liquid in direct contact with its electronics, and one image tends to spring to mind: a motherboard sitting in a tank of oil.
That image has some history behind it, but dielectric cooling itself is hardly a recent experiment. Dielectric fluid for cooling was being used long before enthusiasts started filling PCs with mineral oil. Back in 1985, the Cray-2 supercomputer was cooled by pumping an inert dielectric liquid directly through its circuitry, and dielectric fluids have been insulating and cooling electrical equipment such as transformers for much longer than that.
But what is the fluid actually doing? Why can it come into contact with powered electronics without causing a short circuit? Where does it actually sit, and what happens when a component needs replacing?
To answer that, we first need to tackle a more basic question: what is dielectric fluid?
What is dielectric fluid?
A dielectric fluid is a liquid that does not readily conduct electricity, allowing it to come into direct contact with electronic components while also carrying heat away from them. In practical terms, “dielectric” means electrically insulating: the starting point for direct-contact liquid cooling.
Put tap water onto energized electronics and you have a very obvious conductivity problem but put a dielectric cooling fluid around them and that problem is removed. The fluid can contact the components, pulling heat away without becoming part of the electrical circuit.
That gives the fluid two jobs to do: it needs to interact safely with the electronics while also absorbing and carrying away the heat they produce.
In a precision liquid cooling system, like the systems we develop at Iceotope, that means the dielectric fluid comes into direct contact with the components generating the most heat. Rather than keeping the coolant inside a separate cold plate, our dielectric cooling fluid surrounds the heat source, absorbing the heat and carrying it out of the server to be rejected elsewhere in the cooling system.
But not every form of liquid cooling uses dielectric fluid. The type of coolant used depends on whether it comes into direct contact with the electronics themselves.
What liquid is used in liquid cooling?
There isn’t one standard liquid used for liquid cooling. It all depends on where the coolant actually goes.
In direct-to-chip cooling, the coolant is usually water or a water-glycol mixture. The coolant runs through a cold plate attached to the processor, picks up the heat and carries it away. Crucially, the liquid stays inside that cooling loop so it never comes into contact with the electronics themselves.
Dielectric cooling is a different proposition. The fluid flows directly around the electronic components, which means it needs to be electrically insulating as well as capable of carrying heat away - which is where dielectric cooling fluid comes in.
This is also why there isn't one universal liquid for liquid cooling. Sometimes the coolant stays safely inside a pipe or cold plate, sometimes the coolant is the environment around the components themselves. It all depends on the type of cooling system.
Interested in the differences between precision liquid cooling and direct-to-chip cooling? Take a look at our comparison.
Water works perfectly well when it is kept inside a cold plate but let it flow directly over electronic components and things get complicated very quickly.
What makes dielectric cooling fluid suitable for electronics?
Electrical insulation is only the starting point. A liquid can be non-conductive and still be a terrible choice for cooling high-value computing hardware.
The properties of dielectric fluid need to suit the entire cooling system across its entire lifecycle. That means looking at how well the fluid transfers heat, how easily it can be circulated and how it behaves when it spends years in contact with server materials.
Dielectric strength is one of the basics. The fluid needs to remain electrically insulating at the temperatures and under the operating conditions it will actually encounter.
Then there’s thermal performance. A dielectric fluid for cooling needs to absorb heat from the components and carry it away efficiently. Viscosity is crucial here, because a fluid that is harder to move can increase the pumping effort needed to keep it circulating.
And then there’s material compatibility. Servers contain metals, plastics, seals, cables and other materials that could spend years in contact with the fluid. A coolant can be electrically safe and still cause problems if it’s not compatible with those materials over the long term.
That’s why a suitable dielectric coolant needs to be tested against the hardware around it, rather than simply assumed to be safe because it doesn’t conduct electricity.
Fire behavior and stability come into it too. The fluid needs to remain predictable at higher temperatures and over long periods of use, with its flash point and thermal stability understood before it goes anywhere near a server.
There’s a reason that the liquid dielectric material used for cooling is engineered specifically for the job. Being electrically insulating is essential, but the fluid also needs to perform well thermally and remain compatible with the hardware it is in contact with.
Is dielectric fluid basically mineral oil?
This is where some of the confusion around dielectric cooling comes from.
Back in the 1990s and early 2000s, computer enthusiasts really did experiment with cooling hardware by submerging it in mineral oil. It was cheap, electrically insulating and, at least in principle, capable of carrying heat away from the components.
That history has stuck around, but mineral oil isn’t really the defining feature of dielectric cooling. Some modern dielectric fluids are still hydrocarbon or oil-based. The difference is that they’re engineered and validated for use around modern computing hardware.
That means understanding how the fluid behaves electrically and thermally, whether it’s compatible with the materials inside the server, and how it performs over long periods of use. Fire behavior and handling requirements also need to be understood as part of that picture.
So yes, the mineral-oil experiments are part of the history, but they’re a dated picture of what dielectric cooling looks like today. Once you move beyond the old image, the cooling process itself starts to look very different too.
Single-phase vs two-phase immersion cooling
In single-phase immersion cooling, the dielectric fluid absorbs heat from the hardware while staying liquid. The warmed fluid is then circulated away from the server, where the heat is transferred out of the system before the fluid returns to cool the hardware again.
Two-phase immersion cooling takes a different approach. The fluid boils when it reaches the hotter components, turning into vapor and rising away from the hardware. It then condenses back into liquid and drips back down into the fluid below, using the phase change itself to move heat away.
That means the difference in single-phase vs two-phase immersion cooling comes down to more than the type of fluid being used. The two approaches move heat through the system differently and introduce different requirements around containment, vapor management, fluid loss and servicing.
These are two separate questions. Single-phase and two-phase describe what happens to the fluid as it absorbs heat, not how the hardware is physically arranged. A system can be single-phase without being a tank.
Iceotope uses single-phase dielectric cooling inside a sealed chassis, so the fluid remains liquid as it moves around the components, absorbs their heat and carries it out of the server.
What servicing actually involves
The real test comes when something inside the server needs attention:
- A DIMM fails
- A drive needs replacing
- Someone needs to get into the server, replace the component and get it back into service
It's easy to assume that maintenance means lifting wet hardware out of a tank before any work can begin. With immersion systems, that isn't far off.
Hardware needs to be accessed from the tank and, depending on the system, technicians may need lifting equipment or lift-assists to handle submerged servers. There’s also a much larger volume of fluid around the hardware to deal with during servicing.
Our precision liquid cooling takes a different approach. The dielectric fluid is contained within the server chassis, which remains rack-mountable and front serviceable.
Most components can be serviced much as they would be in a conventional air-cooled server, without draining the dielectric fluid first. Our systems also use removable chassis hoses and slide rails to make in-rack servicing easier.
That way, a failed component doesn’t automatically mean draining a large volume of coolant or moving a fluid-filled server out of the rack. The hardware stays where you would expect it to be, while the cooling system stays contained within the chassis.
How much fluid are we actually talking about?
The two approaches also use very different amounts of fluid. Tank immersion systems surround the hardware with a large volume of dielectric fluid, while a sealed-chassis system only needs enough fluid to circulate within the server.
Our precision liquid cooling systems use 5 to 10 times less fluid than typical single-phase immersion tanks.
That changes more than the amount of coolant being purchased. It means less fluid to contain, manage and handle during servicing, while keeping the cooling system inside a familiar rack-mounted chassis rather than a separate tank.
With precision liquid cooling, the server stays a server. The cooling is happening inside the chassis.
Which concerns about dielectric cooling are actually justified?
If liquid cooling still brings to mind a server sitting in a huge bath of oil, that's understandable. The old mineral-oil experiments have a lot to answer for.
But some of the concerns behind that picture do hold up. The fluid needs to work with the materials inside the server, handle the temperatures involved and remain stable over time. And when something fails, someone still needs to be able to get into the server and replace it without turning maintenance into a major operation.
Others don't. The fluid does need to be electrically insulating, but that’s what the dielectric property is for. And with precision liquid cooling, the dielectric fluid stays inside the chassis and flows directly around the components producing the most heat. The server remains rack-mounted, and most components can be serviced without draining the fluid first.
That’s the thinking behind Iceotope’s precision liquid cooling: use the fluid where it’s needed, keep it contained and make sure the hardware remains practical to run and maintain.
Precision liquid cooling and tank immersion both put fluid in contact with the hardware, but they differ in containment, volume and servicing. Our piece on rethinking tank immersion cooling takes each in turn.
If you’re working out whether precision liquid cooling fits your platform, start a conversation with Iceotope.