A liquid load bank is not a CDU
Original diagram. Not a site record.
A liquid load bank puts heat and flow where the rack will later put heat and flow. That is the job. It is not a coolant distribution unit, and it is not a proof that the IT equipment will boot.
Teams blur those jobs because the pipes look similar. The bank rejects heat the way a loaded rack rejects heat, so the hall-side loop can be run before the racks exist. The CDU is the plant that supplies that loop. If the bank is asked to "be the CDU for a day," the test no longer has a plant and a load. It has one box doing both, and a pass that will not survive the day the real CDU is connected.
The same split applies to the IT. A stable supply temperature at the bank does not mean a GPU tray will train. Boot, optics, and job health are a different sentence. Write that limit on the requirements sheet before anyone cuts metal. A sheet that says "represents the rack" without saying what it does not represent will be quoted as a full-hall proof.
Interface numbers do not belong in a general article, and they do not belong in the book. The specification then in force is cited on your own sheet. A marketing multiple for coolant flow is not that specification. If the only flow number in the room came from a slide, stop and get the document that actually governs the test.
From Design and Build a Liquid Load Bank for NVL72 Racks, Tristan, Data Center Field Guides. No store link until the Kindle edition is listed.
Related: Two documents before you cut metal.
Master the Complete Protocol: Design and Build a Liquid Load Bank for NVL72 Racks
A Field Guide to Interface Boundaries, P&ID Hydraulics, Thermal Envelopes, and Acceptance Sheets
Stop relying on contractor spreadsheets and vendor marketing claims. This field guide provides the complete, auditable verification binder—including physical equations, inspection stations, acceptance thresholds, and signed refusal templates.