Sizing the Secondary Fluid Network: 4", 6", and 8" Header Loops

The header is the backbone of a liquid-cooled hall. Every rack hangs off it, and its size sets three things at once: how much flow the row can carry, how hard the pumps have to work to move that flow, and how much density headroom the design has before it has to be torn out and re-piped. Get it right and the network runs quietly for a decade. Get it wrong and you're either starving racks or fighting noise, erosion, and pumping cost for the life of the facility.

This article walks the logic behind choosing a 4-, 6-, or 8-inch header — and why 4-inch is the common answer.

What the secondary fluid network is

The secondary fluid network (SFN) is the clean-side distribution system between the CDU and the racks. From each coolant distribution unit, supply and return headers run along a row of racks — overhead or under the floor — and each rack taps off both with a drop. It's called the secondary network because it sits on the technology cooling loop, separated by the CDU from the facility water on the primary side, carrying a controlled coolant such as PG25, deionized water, or a water-glycol blend.

The header is the part everything else depends on. Size it for the flow the row will actually draw, at a velocity the piping can live with, within a pressure-drop budget the pumps can meet.

The sizing logic: density drives flow drives diameter

Work it from the load out. Each kilowatt of liquid-cooled IT has to be carried away by a certain amount of coolant. A widely cited industry figure — reflected in OCP's liquid cooling guidance — is roughly 1.5 liters per minute per kilowatt (about 0.4 GPM/kW) at a target coolant temperature rise near 10°C, varying with the coolant and the temperature difference the system is designed around. Warmer facility water narrows the gap between chip and coolant and pushes the required flow per watt up; a bigger allowable temperature rise pulls it down.

Sum that across the racks on a header and you have the row's flow. A row of high-density racks can add up fast, and that total flow — not the diameter of any single drop — is what sets the header size.

Why 4-inch is typical, and when to go bigger

Header diameter is really a velocity decision. For a given flow, a smaller pipe means faster fluid. Push the velocity too high and you get erosion at the turns, noise, and rising pressure drop; run it too low and you're paying for pipe and space you don't need. Designers generally keep header velocity in a moderate band — commonly a few to several feet per second — and size the diameter to land the row's total flow inside it.

For most current deployments, a 4-inch header carries a row's worth of flow comfortably within that band, which is why it's the default. As rack density climbs, as more racks hang off a single header, or as the design reserves headroom for a denser future, the row's total flow rises and the header steps up to 6-inch or 8-inch to hold velocity and pressure drop in check. The choice isn't about the racks individually — it's about the aggregate flow the header has to move and the velocity ceiling it has to respect.

Pressure drop and what adds to it

Velocity sets the baseline, but the pumps see total pressure drop, and the header is only part of it. Run length adds loss. So does every device in the path — valves, fittings, and quick disconnects each take a bite. This is where component choices on the drops feed back into the header design: full-port isolation valves and low-loss transition fittings keep each drop's contribution small, and across dozens of drops that discipline is the difference between a pump curve that fits and one that doesn't.

It's also why the header can't be sized in isolation from the connections hanging off it. The network is one hydraulic system.

Practical routing realities

Diameter interacts with the physical build. Under a raised floor, insulation adds to the effective size — roughly an inch of insulation adds two inches to a wrapped pipe's outside diameter — and headers may have to cross one another, which eats vertical space. Overhead runs free up the floor but load the structure and complicate drops. None of this changes the hydraulic sizing, but all of it constrains how the chosen diameter actually gets installed, and it's worth resolving before the size is locked.

Transitions: adapting across sizes

A real network isn't one diameter end to end. The CDU connections, the header, and the rack drops are all different sizes, and the coolant has to step between them cleanly. That's the job of reducers and transition fittings: bring a 4-, 6-, or 8-inch header down to the line feeding each rack, and reconcile the connection standards between the header, the valves, and the hoses. Every one of those transitions is a potential leak path, so the goal is to make each size change in as few well-made stainless joints as possible.

Design for the density you'll have, not the density you have

Rack power is still climbing, and re-piping a live hall is expensive and disruptive. Sizing a header with some flow headroom — or choosing the next size up where a row is likely to densify — is cheap insurance against a forced upgrade. The header is the hardest part of the network to change later, so it's the part most worth getting right the first time.

Where Titan fits

Titan is a single-source solution provider for the connections these systems depend on — the adapters, fittings, quick disconnects, and valves you have to spec. And because Titan manufactures in the US, it turns orders around with one of the fastest lead times in the industry: weeks, not months.

Titan Fittings supplies the stainless transition fittings and reducers that adapt across 4-, 6-, and 8-inch headers, along with the isolation ball valves, flexible hose connections, and 1-inch OCP LQC quick disconnects that complete each drop. All in 304 and 316L, all compatible with PG25, deionized water, and the coolants these networks run, for the hyperscalers, mechanical contractors, and colocation providers building them.

Tell us your header sizes and rack count and we'll help you spec the transitions and drops. For the connection detail, see the rack drop, component by component, and for the whole system, the full-loop overview.