The Loop Is Running. Is It Ready for the GPUs?

The pump starts. Coolant reaches the farthest branch. Every hose stays dry. Someone looks at the dark racks and asks the only question that matters: can we connect the GPUs?

Not yet.

A first circulation is like starting an engine after a rebuild. Motion is encouraging. It does not prove that every passage is clean, the coolant is correct, trapped gas is gone, or every component will stay within its pressure limit.

ASHRAE Technical Committee 9.9 put this distinction into focus in its May 2026 bulletin, TCS Coolant Integrity and System Readiness Best Practices.

A technology cooling system, or TCS, is the equipment-side loop that carries coolant to the racks. The bulletin turns readiness into four simple questions:

• Coolant: Did the system receive the fluid it was designed to use?
• Air: Can trapped gas leave the real installed pipework?
• Stainless steel: Was the entire wetted path cleaned and protected after fabrication?
• Pressure: Can every component handle every operating and service condition?

Flow proves circulation. It does not prove readiness.

Running is not the same as ready

The coolant supplier sees a formulation. The contractor sees pipe. The controls team sees pressure. The rack vendor sees inlet conditions. The cold plate receives the result of every decision made upstream.

ASHRAE TC 9.9 connects those views. It does not give every facility one universal set of pass limits. The real test is whether the installed loop still matches its approved fluid, materials, pressure and flow model, and operating plan.

1. Did the right coolant go in?

A liquid cooling system is designed around a recipe, not the word coolant. Two clear liquids can look identical in a sight glass while behaving very differently inside a pump or heat exchanger.

A nominal 25 percent by volume propylene glycol and water mixture is common in many single-phase TCS designs, but it is not a universal requirement. The approved formulation, concentration, inhibitor package, and dilution-water quality come from the actual design and equipment requirements.

Change the recipe and viscosity, density, specific heat, and thermal conductivity can all move. That can change pressure drop, branch flow, pump behavior, heat exchanger performance, and control stability. Coolant may reach every rack while the loop no longer behaves like the design model.

Record the design fluid. Control what enters during fill and makeup. Verify the initial charge with an approved laboratory method, then repeat the check after significant additions and on the schedule set for the system.

2. Can trapped air get out?

A new loop starts full of air. Filling pushes most of it out, but high points, inverted bends, flexible hoses, dead legs, and weak vent locations can keep pockets behind.

An air pocket behaves like a valve nobody installed. It can restrict a branch, weaken heat transfer, create pump noise or cavitation, and make startup results change from one run to the next.

Temperature and pressure matter too. Gas dissolved in a cold fill can form bubbles as temperature rises or pressure falls. That is why warm circulation can reveal a pocket that a cold flush missed.

Vents, separators, purge connections, hose routing, and functional tests must work in the installed loop. Their job is simple: give air a reliable way out without exceeding equipment limits.

3. Was every wetted surface cleaned and protected?

Stainless steel resists corrosion with a very thin chromium-rich oxide film. Think of it as a molecular raincoat. Welding, grinding, handling, and fabrication can leave heat tint or free iron that weakens that protection.

Approved cleaning removes fabrication residue. Passivation helps restore a suitable corrosion-resistant surface. Together, those steps reduce early metal release, inhibitor demand, particle formation, and fouling risk downstream.

Most is not enough. One late weld, untreated replacement part, dirty storage period, or long stagnant delay can put contamination back into the wetted path.

4. Can every component handle every pressure case?

Normal operation is only one pressure case. Checking only the normal pump point is like evaluating a bridge only on a windless day. The condition everyone overlooks may set the real limit.

The loop also experiences fill, cleaning, flushing, static height, expansion tank precharge, relief settings, pump edge cases, and short pressure changes while equipment is connected or disconnected.

A pressure rating is a ceiling, not a target. A pump can have enough head to move coolant while a hose, quick disconnect, cold plate, or another component is too close to its limit.

A hydraulic model predicts pressure and flow through the whole system. It must cover normal and non-normal states. Controls, alarms, bypasses, pressure-reducing devices, and mechanical relief protection must agree with that model and remain correctly set after handover.

What a real green light looks like

A sight glass can look perfect while an air pocket sits beyond it. Stable flow can coexist with the wrong formulation. A passed pressure test says nothing about whether a field weld was cleaned and passivated. No single signal earns permission to connect valuable hardware.

Imagine a cold flush that reaches the expected flow. The team warms the loop, dissolved gas forms bubbles, and a high point starts collecting air. The first flow result was not wrong. It was incomplete because the operating condition changed.

Before connection, the next operator should be able to answer five questions without starting a document hunt:

1. Which coolant was approved, and what actually entered? Keep the formulation, concentration, water quality, mixing record, and laboratory result together.
2. Where can air collect, and how does it leave? Document high points, vents, separators, purge points, hose routing, and stable behavior after warm circulation.
3. Was the full stainless path cleaned and treated? Include late welds, replacement parts, storage protection, verification records, and the approved quality plan.
4. Did the pressure review cover every important state? Include component ratings, fill and flush conditions, static height, expansion settings, controls, and mechanical safeties.
5. Who can approve the connection? Set the acceptance criteria, corrective actions, signoffs, and final authority before the racks are waiting.

The loop is ready for site review when those records agree. The site team still makes the final connection decision.

The baseline starts at commissioning

Commissioning creates the first trusted picture of normal behavior. It should not be the last time anyone checks whether the original assumptions still hold.

Racks are added. A stagnant branch starts. A filter is changed. A hose is disconnected. Makeup fluid enters. Pipework is modified. Each event can change fluid condition, particles, air, flow, or pressure.

A sample bottle is a photograph. It can be sharp and useful, but it cannot show what happened before or after the shutter clicked. A living baseline adds the timeline around the sample: fluid condition, pressure, flow, filtration, pump state, temperature, makeup, maintenance, and workload.

How continuous monitoring helps

Reliability Engine brings commissioning records and ongoing fluid data onto one timeline. Teams can compare fluid condition, flow, pressure, filtration, makeup, samples, alarms, and service events with the approved project criteria.

• Before connection, continuous monitoring shows whether fluid, flow, filtration, and air signals are settling together or telling different stories.
• During operation, predictive analysis links a change to load, makeup, service, or another recorded event so unusual drift is easier to investigate.
• After maintenance or expansion, the same record shows whether the loop returned to known behavior or established a new condition that still needs review.

Reliability Engine complements approved laboratory methods, equipment limits, mechanical safeties, controls, codes, site procedures, and commissioning authority. The site team keeps the final connection decision. The platform helps the team see the records and trends together instead of comparing disconnected snapshots.

Commissioning establishes known-good behavior. Continuous monitoring shows when the loop leaves it.

The simple rule

ASHRAE TC 9.9 does not ask every cooling loop to use the same fluid, instruments, or limits. It asks each team to protect the design it actually approved.

Coolant reaching the farthest branch is the start of commissioning, not the finish. A loop is ready for connection review when the coolant matches the design, air can leave, wetted surfaces are protected, pressure stays within every applicable rating, safeguards work, and the site authority accepts the result.

Bringing a direct-to-chip cooling loop online? Talk to Reliability Engine about building a living readiness record.

References

1. ASHRAE TC 9.9: TCS Coolant Integrity and System Readiness Best Practices, May 2026
2. ASHRAE Technical Committee 9.9: Mission Critical Facilities, Data Centers, Technology Spaces and Electronic Equipment
3. ASHRAE TC 9.9 Datacom Encyclopedia

The ASHRAE publication is informational guidance, not a standard. It does not supersede IT equipment manufacturer requirements, applicable codes, or site-specific operating procedures.