Rack manifold valve tags before cold-plate connection

Direct answer

Before cold-plate connection, a rack manifold valve tag and flow-direction photo record should identify the data hall, row, rack, cabinet side, manifold set, CDU or row loop, supply manifold, return manifold, valve tag, valve service, valve normal position, actual position, lock or cap status, flow-direction arrow, hose label, hose route, quick-disconnect type, cold-plate connector ID, drip tray or containment, leak sensor zone, BMS or rack-manager point, pressure or hold status, flushing and cleanliness status, open exceptions, witness, and the exact connection released.

The record should prove that supply is not being confused with return, red and blue or project color conventions match the installed manifold, valves are tagged and in the intended state, hose ends are labeled for flow direction, hoses are not kinked or rubbing, quick disconnects are compatible and locked, leak detection is in place, and the TCS loop has passed the project-required cleaning, flushing, pressure, and leak checks before any server cold plates are connected.

Use this field note as documentation guidance only. The rack OEM, cold-plate OEM, CDU manufacturer, manifold manufacturer, project specifications, commissioning script, water-treatment plan, BMS sequence, site safety plan, electrical lockout rules, leak-response plan, and owner acceptance process control the actual connection, valve operation, pressure limits, coolant chemistry, and traffic in the white space.

Why this hold point matters

Direct-to-chip liquid cooling makes the mechanical record part of the IT release path. NVIDIA's DGX GB rack documentation describes liquid running through rack manifolds and then through cold plates attached to CPUs and GPUs. Lenovo describes a water manifold delivering water to enclosure inlet and outlet connectors. Panduit describes a rack manifold with supply and return pipes for liquid-cooled servers.

That means a mislabeled valve, reversed hose, hidden kink, unverified quick disconnect, or unready leak sensor is not just a piping punch item. It can affect cold-plate flow, leak response, service access, and whether the rack can be safely commissioned.

The strongest record is made before connection, while tags, arrows, caps, valves, hoses, and sensors are still visible. After cold plates are connected and cable bundles close the rear of the rack, the evidence gets harder to collect.

Start with the connection basis

Open the packet with the controlling documents. Include the rack elevation, manifold drawing, CDU or row manifold drawing, P and ID, valve schedule, hose schedule, cold-plate connection map, OEM installation guide, commissioning script, water-treatment requirement, flushing record, pressure or leak-test record, leak-detection drawing, BMS point list, safety plan, and the person authorized to release the connection.

Do not use a generic supply-return convention as the release basis. Vertiv, Lenovo, Panduit, CoolIT, NVIDIA, and other sources reviewed for this package show different rack formats, feed locations, connector arrangements, color conventions, service kits, leak detection approaches, and CDU layouts. The project documents and OEM instructions decide what must be tagged and photographed.

If the project has a formal hold point, name it. Examples include rack manifold ready for hose connection, CDU secondary loop ready for rack connection, leak detection ready, pressure test complete, cold-plate connection released, or first fill with IT equipment connected.

Separate FWS, TCS, and rack records

A clear record separates the facility water system, the technology cooling system, the CDU, the row or room manifold, the rack manifold, and the server cold-plate loop. ASHRAE describes the CDU as a common way to separate the facility water system from the technology cooling system while providing heat transfer, pressure control, filtration, and coolant-temperature control.

Vertiv's leak detection white paper defines the technology cooling system as the liquid circuit from the CDU to the rack, through the manifold and IT equipment, and then back through the return manifold to the CDU. That definition is useful in the field because it tells the record where the rack manifold sits in the loop.

Do not let labels drift between loops. A primary-side facility valve, a secondary-side TCS valve, a rack branch valve, a manifold isolation valve, a drain valve, and a cold-plate quick disconnect are different objects. Tag and photograph them as different objects.

Map supply and return at the rack

Photograph the rack rear before hoses are connected. Capture the rack ID, row, cabinet side, manifold pair, feed direction, top or bottom feed, supply side, return side, port numbers, unused caps, drain point, vent point, pressure gauge where present, and the physical clearance around power shelves, PDUs, cable cartridges, and network cabling.

Vertiv's in-rack manifold manual describes configurations where return is red coded and supply is blue coded, and it instructs installers to mount the return and supply manifolds on specified rack sides for that configuration. Treat that as product-specific evidence, not a universal color rule. If the project uses another convention, photograph that convention and the drawing that controls it.

The map should make the connection review possible without standing at the rack. A reviewer should be able to say which manifold is supply, which is return, which valve isolates each branch, and which cold-plate connector each hose is meant to serve.

Valve tag and position photos

Every valve that can affect the cold-plate connection should have a visible tag or another project-approved identifier. Photograph the valve tag, valve body, handle, normal position, actual position, flow arrow if present, lockout or car-seal status where required, cap or plug status, and the connected line label.

Include the valve schedule in the packet. The field note should tie each tag to service: rack supply isolation, rack return isolation, manifold drain, manifold vent, pressure relief, bypass, fill, CDU secondary supply, CDU secondary return, automatic shutoff, energy valve, or project-specific control valve.

Do not assume an open handle means the valve is correct. Record whether the connection hold point requires valves open, closed, locked, capped, vented, depressurized, or under nitrogen shipping pressure. Vertiv's in-rack manifold manual includes depressurizing and drain-valve handling steps for its product. Other products may differ.

Flow direction and hose labels

The flow-direction record should show both sides of the connection: manifold label and hose label. Photograph the supply and return markings, arrow labels, hose-end labels, color bands, tags, QR labels, port numbers, and cold-plate connector identifiers before mating quick disconnects.

Vertiv's CDU 100 commissioning guide includes a pre-connection check that all hose ends are correctly labelled for supply/return flow direction, that hoses run smoothly from CDU to racks without kinking, and that final hose connections to the CDU and racks are tight. A technical cooling article reviewed for this package also calls for clear labeling, including flow direction, for serviceable components.

A useful record states what was checked, not only that labels exist. Supply hose S-R12-08 was matched to Rack R12 supply manifold port S08 and cold-plate inlet CP-08 is stronger than hoses labeled.

Cold-plate connector match

Before connecting to cold plates, record the connector match. Include rack port number, server tray or node ID, cold-plate inlet, cold-plate outlet, connector gender or profile where applicable, quick-disconnect model where required, cap condition, seal or O-ring condition if visible and inspection is allowed, cleanliness, and whether any adapter is used.

CoolIT describes rack manifolds connecting CDUs to direct-liquid-cooled servers and offering dry-break quick disconnect options. Panduit describes rack manifold inlets and outlets with leak-free quick connectors for rapid connection and disconnection. DCD's direct liquid cooling white paper warns that quick-connect couplings between IT and manifold must be mechanically compatible.

Do not force an uncertain connector. If the connection is not aligned, if the latch does not seat, if the cap is missing, if the seal is damaged, if the hose is mislabeled, or if the inlet and outlet map is unclear, hold the connection and record the exception.

Hose routing and bend evidence

Photograph the hose route before final connection and after connection. Capture the full route from rack manifold to cold plate or from CDU to rack where relevant, bend radius, strain relief, clamp or bracket support, contact points, cable crossings, door swing, service loop, removable module path, and any rub point against bus bar covers, PDUs, cable managers, tray rails, or sharp sheet metal.

Vertiv's in-rack manifold manual warns against crushing or kinking hoses and says to limit curvature to the minimum bend radius recommended by the manufacturer while avoiding sharp bends at end fittings and manifold connections. DCD's white paper similarly states that flexible cooling pipes should not exceed bend-radius limits because doing so can increase leak risk and impede coolant flow.

The photo record should include rejected routing. A hose that looked acceptable before the server tray was seated may kink when the tray moves into final position. Keep the failed photo and retest photo so the release explains what changed.

Cleanliness and flushing status

Cold plates have small passages, so cleanliness is part of the connection decision. Record the TCS flushing status, filter status, strainer status, water or coolant batch, water-quality checks required by the project, temporary bypass loops, caps, plugs, open-end protection, and whether construction activities that create debris are still active near the loop.

Vertiv's single-phase TCS deployment guide says surface finish and cleanliness of pipes, valves, and fittings are critical because of small openings inside server cold plates. It also says CDUs should be connected to the TCS only after initial TCS flushing processes are complete. Lenovo's Neptune standards emphasize build clean, ship clean, install clean, and maintain clean, including pH, conductivity, bacteria count, and inhibitor concentration monitoring.

Do not connect cold plates to a dirty or uncertain loop just because the mechanical work looks finished. If flushing records, filter changes, fluid labels, water-treatment signoff, or open-end cap status are missing, mark the connection hold.

Pressure, vent, drain, and state

Before mating connectors, state the pressure condition. Record whether the rack manifold is under shipping pressure, depressurized, filled, isolated, vented, drained, flushing, pressure-tested, or in service. Photograph pressure gauges, test ports, vent points, drain valves, caps, and any hold tag.

Vertiv's leak detection white paper discusses gas pressure dwell testing, hydrostatic pressure testing, active testing, and pressure monitoring as leak-test approaches. Vertiv's in-rack manifold manual also notes that the product may arrive pressurized with nitrogen and must be vented through the drain valve before installation or movement for that product.

Do not generalize pressure numbers. Test pressure, test medium, dwell time, acceptance, valve state, and whether quick disconnects are open or closed are project- and product-controlled.

Leak detection and containment readiness

The connection record should prove leak detection is ready before liquid is introduced to IT equipment. Photograph drip trays, leak tape, leak ropes, point detectors, rack sensors, tray sensors, BMS point labels, alarm address, cable routing, sensor dry condition, and any containment pan or drain.

NVIDIA describes node or tray, rack, and data-center liquid leak detection levels for DGX NVL72 systems, including cold-plate leak sensors, inner manifold leak sensors, rack-level sensing ropes, and CDU or piping sensors. Vertiv recommends monitored leak detection for units and supply and return lines, and its leak detection white paper says leak detection should be placed near connections where risk is highest.

A leak sensor photo is not enough if the point is not mapped. The packet should say which sensor covers Rack R12 manifold supply ports, which point appears in BMS or rack management, what alarm text is expected, and whether the leak-response sequence is manual or automatic.

Controls and valve-response boundaries

Record what the controls are allowed to do. Some systems only alarm. Some close supply and return valves. Some notify the rack manager, CDU, BMS, or operations team. Some power down trays or racks under defined leak states. The cold-plate connection release should not guess.

NVIDIA's BMS integration document describes rack-level leak detection responses that include shutting off liquid valves for rack supply and return and notifying management systems. Vertiv's deployment guide recommends monitored fluid detection wired to automatic closure of field-installed supply and return shutoff valves in some cases. DCD's white paper notes that owners must decide how much control is given to automatic intervention compared with notifications.

Keep this section factual. Record the BMS points, valve tags, alarm test status, control sequence version, and response boundary. Do not claim an automatic shutdown or valve closure exists unless it has been configured, tested, and accepted for this rack.

Use an auditable connection table

Use the OEM checklist, commissioning script, or owner form first. Add a table where those forms do not tie tags, flow direction, valve state, hose evidence, leak detection, and connection release together.

Build the photo packet

A useful packet starts wide and then moves close. Include rack rear overview, rack ID, row label, manifold pair, valve tags, valve handles, supply and return labels, flow arrows, hose-end labels, quick-disconnect caps, cold-plate connector IDs, hose route, bend radius, leak detection, drip tray, pressure gauge, drain or vent, BMS point screen where allowed, and final connection release photos.

Use scale and orientation. A photo showing Rack R12, rear-left supply manifold, port S08, and hose tag S-R12-08 is more useful than a close-up of a blue hose with no context.

Name or caption photos with rack, side, manifold, port, valve tag, state, and status. A later leak review should be able to trace Valve RM-R12-S-ISO from drawing to field tag to hose connection to BMS point.

Before cold-plate connection checklist

Run this check before representing rack manifold valves and flow direction as ready for cold-plate connection.

• Confirm the release basis: rack drawing, manifold drawing, P and ID, valve schedule, hose schedule, OEM guide, commissioning script, leak-detection drawing, and approval authority.
• Confirm rack identity: data hall, row, rack, cabinet side, CDU or row loop, manifold set, and server tray range.
• Photograph supply and return manifolds with color convention, flow arrows, port numbering, top or bottom feed, and unused caps.
• Photograph every relevant valve tag, valve service, handle position, normal position, lock or cap status, drain, vent, bypass, and automatic shutoff where present.
• Match each hose label to the manifold port and cold-plate inlet or outlet before mating quick disconnects.
• Confirm hose ends are labeled for supply or return flow direction and routed without kinks, rub points, sharp bends, or service-access conflicts.
• Confirm quick disconnects are compatible, clean, capped until connection, aligned, and locked after connection.
• Confirm flushing, filtration, water-quality, pressure-test, and leak-test records required by the project are complete before connecting IT equipment.
• Confirm pressure state: shipping pressure, depressurized, filled, vented, isolated, or in service, with test medium and acceptance basis recorded.
• Confirm leak detection: sensor zone, rope or tape route, point detector, drip tray, BMS or rack-manager point, alarm test, and response sequence.
• Preserve exceptions such as missing labels, reversed tags, kinked hoses, unclear valve state, untested alarm points, dirty caps, damaged connectors, or missing pressure records.
• State the exact cold-plate ports released and the exact ports still held.

Failed conditions and holds

Do not clean up the record by deleting failed first checks. The failed check often explains why the rack was held and what was corrected before connection.

Common holds include unlabeled supply or return hose ends, missing flow arrows, valve tag mismatch, handle position not matching the commissioning script, reversed red and blue convention, hose kink at a manifold connection, bend radius conflict when the tray is seated, incompatible quick disconnect, missing cap, dirty connector, unverified pressure state, no flushing signoff, no leak-detection point, alarm text not matching the BMS point list, and wet or suspect drip tray before connection.

The retest should say what changed, which photo proves it, which party accepted it, and whether the cold-plate connection was released, partially released, or still held.

Weak and strong records

Weak note: Rack manifold labeled and ready. Hoses connected.

That note does not identify the rack, manifold pair, valve tags, supply-return convention, flow direction, hose labels, cold-plate connector IDs, valve state, pressure state, flushing record, leak detection, failed checks, or release boundary.

Stronger note: Rack R12 in Data Hall 2, Row C was inspected for cold-plate connection release on 2026-06-09. Basis was the approved rack manifold drawing RM-R12-4, valve schedule V-9, Vertiv in-rack manifold manual, CDU 100 commissioning script, water-treatment signoff WT-14, and leak-detection drawing LD-C-R12. Rear overview photos show the return manifold on the rear-right side under the project red-return convention and the supply manifold on the rear-left side under the project blue-supply convention.

Valve photos show RM-R12-S-ISO and RM-R12-R-ISO tagged, capped, and closed for the pre-connection review. Flow-direction photos show hose tags S-R12-01 through S-R12-12 and R-R12-01 through R-R12-12 matched to manifold ports and cold-plate inlet and outlet labels. Hoses S-R12-07 and R-R12-07 failed the first route check because the bend tightened when Tray 7 was seated. The hoses were re-routed through the upper guide, photographed again, and accepted by the commissioning lead.

The TCS flushing record, filter change record, pressure test, and leak tape installation were attached. Leak rope zone LD-C-R12-REAR and point detector LD-C-R12-PAN were dry and alarm-tested to the BMS point list. Cold-plate connection was released for trays 1 through 6 and 8 through 12 at 14:20. Tray 7 remained held until the OEM service lead approved the revised hose route.

The stronger note works because it connects rack identity, valve tags, supply-return direction, hose labels, connection status, failed routing, correction, leak detection, and release limits.

Questions that come up

Is red always return and blue always supply? No. Vertiv uses that convention in the cited in-rack manifold manual, but the project drawing and OEM instructions control the actual color convention.

Can cold plates be connected if the rack manifold pressure test passed? Only if all other required holds are released. Pressure testing is not the same as supply-return labeling, hose routing, flushing, leak detection, BMS point mapping, or connector compatibility.

Should leak detection be verified before or after connection? The safe record verifies it before introducing liquid to IT equipment and then documents any post-connection alarm or dry-tray check required by the commissioning script.

What if a hose label conflicts with a manifold label? Stop and resolve the conflict through the project authority. Do not connect based on color alone.

Can one photo prove the rack is ready? No. The packet needs overview photos, tag photos, valve-position photos, flow-direction photos, hose-route photos, connector photos, leak-detection photos, and release notes.

Compliance and safety limits

This field note is not a liquid-cooling design, cold-plate installation manual, pressure-test procedure, water-treatment plan, BMS sequence, leak-response plan, lockout procedure, electrical safety plan, warranty decision, or OEM acceptance. The rack OEM, cold-plate OEM, CDU manufacturer, manifold manufacturer, approved drawings, specifications, commissioning authority, water-treatment provider, BMS integrator, owner operations team, and site safety plan control the work.

Do not use this checklist to bypass qualified mechanical personnel, OEM service procedures, electrical lockout, stored-pressure precautions, coolant handling rules, PPE, drip containment, leak detection, alarm testing, water-quality limits, pressure limits, quick-disconnect instructions, torque or latch instructions, working-clearance rules, energized equipment restrictions, or owner acceptance. The packet preserves rack manifold valve tag, flow direction, and connection evidence. It does not authorize unsafe cold-plate connection by itself.

Sources checked

• ASHRAE TC 9.9, Water-Cooled Servers Common Designs, Components, and ProcessesUsed for background on water-cooled server components, cold plates, manifolds, valves, CDUs, FWS/TCS separation, water quality, filtration, and service concerns.
• Vertiv, Guidelines for Deployment of Single Phase Technology Cooling SystemsUsed for TCS installation limits, cleanliness, flushing, supply and return leak detection, shutoff valves, pressure, filters, air vents, and commissioning precautions.
• Vertiv, CoolChip CDU 100 Installation and Commissioning GuideUsed for supply and return isolation valves, hose flushing, pressure testing, leak detection tape, hose length, supply/return flow-direction labeling, hose routing, and final connection checks.
• Vertiv, CoolChip 1-Phase Fluid Network In-Rack Manifold User ManualUsed for in-rack manifold configuration, red return and blue supply examples, supply and return hose matching, quick disconnect compatibility, hose handling, drain valve, pressure, and leak detection cautions.
• NVIDIA, DGX GB Rack Scale Systems User Guide, HardwareUsed for rack-scale context showing liquid cooling manifolds, rear access to inlets and outlets, and coolant running through manifolds and cold plates attached to CPUs and GPUs.
• NVIDIA, Base Command Manager Integration with Building Management SystemUsed for leak detection levels, rack and tray sensors, CDU and piping sensors, BMS/BCM integration, and examples of valve shutoff and notification behavior.
• CoolIT Systems, Rack ManifoldsUsed for rack manifold purpose, CDU-to-server connection, top or bottom feed, front or rear rack placement, quick disconnect options, low pressure drop, and balanced flow context.
• Electronics Cooling, Direct Liquid Cooling for High-Compute ServersUsed for serviceability context on quick disconnects, hose length, avoiding kinks, leak detection and containment, cable access, and clear labeling including flow direction.
• Data Center Dynamics and Schneider Electric, Direct Liquid Cooling System Challenges in Data CentersUsed for DLC system architecture, TCS scope, cold plates, quick-connect compatibility, hose bend radius, pressure drops, leak detection decisions, energy valves, and commissioning records.
• Lenovo Press, ThinkSystem SD665-N V3 Neptune DWC ServerUsed for a current rack/server example with rear water manifolds, manifold supply and return, Eaton ball valves, drip sensors, CDU connection, cold plates, and water-quality context.
• Panduit, Direct-to-Chip Liquid CoolingUsed for rack manifold details including supply and return pipes, quick connectors, air exhaust valves, cooling distribution, server supply and return connections, and deployment-error reduction.
• Vertiv, ACS Cold Plate Leak Detection and Intervention White PaperUsed for TCS definition, leak detection near connections, manual and automatic reaction boundaries, pressure testing, dry nitrogen shipping and storage, point detectors, ropes, and leak-location mapping.
• Lenovo Press, Lenovo Neptune Direct Water-Cooling StandardsUsed for broader water-cooling standards context on cold plates, FWS/TCS loops, build clean, ship clean, install clean, maintain clean, water quality, materials, AHJ limits, and loop maintenance.

Related guides