Three quantities determine whether a loop works: how much heat, how much flow, and at what temperature. They are joined by one equation, and it is worth carrying in your head because it settles most design arguments in a single line.

For water, specific heat capacity is about 4.18 kJ/kg·K, and one litre per minute is one kilogram per minute. So:

Heat removed (kW) = flow (L/min) × ΔT (K) × 0.0697

Which rearranges to the form you use on site: one litre per minute removes about 0.07 kW for every kelvin of temperature rise.

Work an NVL72 rack through it. At 142 kW and a 12 K rise across the rack, required flow is 142 ÷ (0.0697 × 12) ≈ 170 L/min, roughly 45 US gallons per minute, into a single cabinet. Widen the design ΔT to 15 K and the same heat needs 136 L/min; narrow it to 8 K and it needs 255 L/min. That trade is the whole of secondary-loop design: higher ΔT buys smaller pipes, smaller pumps and less parasitic power, and it costs you approach temperature at the cold plate.

Supply temperature is a facility decision with a silicon constraint. ASHRAE's liquid classes are named for their maximum supply temperature — W17, W27, W32, W40, W45 — and the higher classes are the interesting ones, because a loop that runs at 32–45 °C supply can reject heat to ambient air through dry coolers for most of the year in most climates, and never start a compressor. That is where the operating-cost argument for liquid actually lives; it is not the pumps, it is the chillerless hours. The constraint is that every degree of supply temperature is a degree you have spent before the coolant reaches the die, so the vendor's maximum inlet specification for the tray is the number that caps your ambition, not the chiller's capability.

Pressure and approach. Cold plates and quick disconnects each contribute pressure drop; a rack's full hydraulic path — manifold, hoses, UQDs, microchannels — typically lands somewhere between 1 and 2 bar at design flow. Secondary pumps are sized for that plus the header run, and the CDU regulates to a differential pressure across the manifold rather than to a flow number, because flow redistributes as trays are added and removed. If you take a tray out and cap the port, the remaining trays get more flow, not less; if you take a tray out and leave the port open, you have a leak.