GPU AIO Liquid Cooler Pump Failure or Blockage? How to Tell a Dead Pump, Airlock & Clogged Cold Plate from Other Overheating

2026-10-23 ｜ GPU Repair ｜ About 15 minutes ｜ HK GPU Fix Editorial Team

Summary

An AIO cooler can fail in ways that all look like overheating, but a dead pump, trapped air, and a clogged cold plate leave different clues. The key distinction is the relationship between coolant temperature and GPU core temperature, with pump speed, sound, and tube temperatures. This guide explains how to read those clues and decide what to check next.

How an AIO Loop Moves Heat, and Why That Matters for Diagnosis

An all-in-one (AIO) liquid cooler is a sealed loop with four working parts. A pump drives coolant around the loop. A cold plate sits against the graphics processor and absorbs its heat into the coolant. Tubes carry the coolant to a radiator. Fans on the radiator blow air through its fins and release the heat out of the case. Nothing in that chain works on its own: heat must be picked up at the cold plate, carried by moving coolant, and released at the radiator. If any one step stops, the core temperature rises, even if the other steps look normal.

That chain is why "the card overheats" is not a diagnosis on its own. A radiator fan that has stopped, a pump that has stopped, a cold plate that is clogged, and a cold plate that is not making proper contact can all produce a high core temperature reading. They differ in what happens to the coolant, and that difference is the most useful diagnostic signal available without opening anything.

Two temperatures therefore have to be read together, not one on its own. The first is the GPU core temperature, which is the temperature of the chip itself. The second is the coolant temperature, sometimes labelled liquid temperature or water temperature in monitoring software, where the cooler reports it at all. Not every AIO reports coolant temperature, and this guide explains what to do when it does not. But when both readings exist, their relationship separates the failure modes more reliably than either reading alone.

Think of it in plain terms. If coolant is flowing and the radiator is working, sustained load warms the coolant gradually, and the core temperature sits above the coolant temperature by a gap that grows with load and shrinks again as the loop cools down after load stops. If coolant is not flowing, the core heats up very fast under load while the coolant stays close to its idle temperature, because the heat never leaves the cold plate area to warm the rest of the loop. If coolant is flowing but the cold plate is partly blocked, the coolant does warm up, but the core runs hotter than expected for that coolant temperature, because heat is not transferring efficiently from the chip into the coolant stream. Those three patterns — cool loop with a hot core, a loop that warms normally but a core that still runs hot, and a loop that warms up together with the core — point in three different directions, and confusing them leads to replacing the wrong part.

Dead Pump: What It Looks Like, and How to Confirm It

A dead pump means coolant is not being driven around the loop at all. The most direct evidence is the pump speed reading. Many AIO pumps report a speed, often on a fan header or through a USB connection to the manufacturer's software. A reading of zero, or a pump entry that has disappeared entirely from monitoring software, is strong evidence — provided the reading is checked in more than one place. A single program can misread a header, so cross-check in the system firmware setup screen or a second monitoring tool before concluding anything. Also check the simple physical causes first: a pump power cable that has come loose, a header that was changed during other work inside the case, or a fan curve that was set to stop the pump at low temperatures. A pump that is told not to run is not a broken pump, and that distinction costs nothing to check.

Sound and touch add a second layer of evidence. A working pump usually produces a faint vibration or a low hum that can be felt or heard when a hand is placed lightly on the pump housing or the tubes near it, especially at startup when the pump spins up. A dead pump produces nothing: no vibration, no change in sound when the pump speed setting is raised, and no flow noise. Be careful with this test, because some pumps are genuinely very quiet, and case fans can mask pump noise entirely. Absence of audible noise alone never proves a dead pump; it is only supporting evidence alongside the speed reading and the temperature pattern.

The temperature pattern is the decisive part. With a dead pump, the core temperature climbs rapidly once load is applied — far faster than it would with a working loop — and it may hit its thermal limit and throttle within a short period of load, while the coolant temperature, if reported, barely moves. The radiator also stays cool to the touch, and the two tubes feel similar in temperature rather than showing the warm-out and cooler-return difference of a flowing loop, because no warm coolant is reaching the radiator at all. If the fans on the radiator are spinning normally and the radiator is cool while the core is extremely hot, the problem is upstream of the radiator: heat is not being carried there. That combination — normal fans, cool radiator, fast-climbing core, zero or missing pump speed — points at the pump or its power and control, not at the fans or the radiator fins.

One caution matters here. Do not keep running heavy load on a card with a suspected dead pump in order to "see how hot it gets." Thermal throttling protects the chip for a time, but repeated sustained overheating is not a diagnostic tool. Once the pattern is clear — fast core rise, no pump speed, cool loop — stop the load, power down, and check connections and settings with the machine off or at idle. If the pump still does not run after its power and control have been verified, the unit needs workshop inspection rather than further load testing.

Airlock: Trapped Air at the Pump, and Why Position Changes the Symptoms

An airlock is trapped air sitting where coolant should be, most troublesomely at the pump itself. A small amount of air exists in most sealed loops over time, because coolant slowly permeates through the tubes and the loop is not refilled. That air normally collects at the highest point of the loop. Trouble starts when the highest point is the pump, or when a bubble is drawn into the pump inlet. A pump designed to move liquid moves air very poorly: it cavitates, churns, and moves little or no coolant, even though it is spinning and reporting a normal speed. That is the trap in airlock diagnosis — the pump speed reading can look completely normal while flow is badly reduced.

The characteristic symptom of an airlock is sound that changes with position and time. Gurgling, bubbling, trickling, or a rattling churn from the pump area, especially for a period after startup or after the machine has been moved, suggests air passing through the pump. The sound may come and go, and it may change or temporarily disappear when the case is tilted gently or the radiator is repositioned so that the pump is no longer the highest point. A second characteristic is inconsistency: temperatures that are acceptable in one session and poor in the next, or a card that behaves differently after the machine has been moved, can both reflect a bubble that has shifted position. A dead pump does not behave this way; it is consistently dead. A clog does not behave this way either; it is consistently restrictive. Intermittent, position-sensitive symptoms point toward air.

Temperature readings in an airlock are messier than in a dead pump, and that messiness is itself a clue. Coolant temperature may rise more slowly than expected, or unevenly, because flow is intermittent. The core temperature may spike during the moments when the bubble blocks flow and recover when the bubble shifts. Tube temperatures may feel uneven or change over minutes rather than settling into a stable warm-out, cooler-return pattern. If monitoring software shows a normal pump speed but the temperatures behave erratically and the pump area gurgles, air is a more likely explanation than a dead pump or a clog.

There are limits to what a user should do about a suspected airlock. Gentle repositioning — running the machine briefly with the radiator positioned so its top is above the pump, so air can migrate to the radiator where it does less harm — is a reasonable, low-risk step, and it is also a diagnostic test: if symptoms clearly improve when the pump is no longer the highest point, air was likely involved. Aggressive shaking, repeated tilting while under load, or attempting to open a sealed loop to bleed or refill it are not reasonable steps. Sealed AIO units are not designed to be opened by the user, and opening one risks coolant loss and damage that turns a manageable problem into a dead cooler. If repositioning does not produce a lasting improvement, or if gurgling returns persistently, the loop needs professional assessment rather than repeated home attempts.

Radiator orientation deserves a clear statement because it is the most common preventable cause. In a typical installation, the radiator should be mounted so that air collects in the radiator, not in the pump. That generally means the radiator's highest point sits above the pump, and, for a vertically mounted radiator, the tube connections sit at the bottom where possible so that any air stays trapped at the top of the radiator away from the tubes. A radiator mounted below the pump, with the pump as the highest point in the loop, invites air to gather exactly where it causes the most trouble. Orientation alone does not prove an airlock, but an installation where the pump is the highest point plus gurgling plus erratic temperatures is a coherent picture, and correcting the orientation is the first thing to try.

Clogged Cold Plate and Degraded Coolant: The Slow, Gradual Failure

A clogged cold plate is a different kind of failure from the two above: it develops gradually, not suddenly. The cold plate contains fine internal channels or fins that the coolant flows through. Over a long service life, sediment, corrosion products, or biological growth in the coolant can narrow those channels. Flow is not stopped outright, as with a dead pump, but it is restricted, and heat transfer from the chip into the coolant becomes less efficient. The result is a slow deterioration: temperatures that were acceptable months ago creep upward under the same load, the card throttles sooner than it used to, and the change is noticed over weeks rather than appearing overnight.

The temperature pattern reflects that gradual restriction. Unlike a dead pump, the coolant does warm up under load, because some coolant is still moving. Unlike a healthy loop, the core temperature runs higher than expected for that coolant temperature, and the gap between core and coolant temperatures widens over time. The radiator may feel warm, and the fans may be working hard, yet the core stays hot — because the bottleneck is at the pickup point, not at the release point. Tube temperatures can also help: a restricted cold plate may show a smaller-than-expected temperature difference between the two tubes, or one tube that warms much more slowly, reflecting reduced flow through the block. These are qualitative observations, not precise measurements, and they should be read as a pattern across several sessions, not from a single reading.

The key diagnostic distinction is the timeline. A sudden failure after the machine was moved, or a sudden failure with gurgling, argues against a clog. A slow, steady worsening over months, with a pump speed that still reads normally and fans that still work, argues for a restriction or coolant degradation. A clog also does not respond to radiator repositioning the way an airlock does: tilting the case does not clear sediment from fine channels, and if symptoms are unchanged by orientation changes, that stability itself argues against air and toward restriction.

There is one common mistake to avoid here. A clogged cold plate cannot be fixed by turning the fans up, replacing thermal paste on the chip, or reseating the cooler, unless the paste or mounting is independently faulty. Those steps address different layers — heat release, and heat transfer from chip to cold plate surface — while a clog sits inside the cold plate, between the surface and the coolant stream. If fan speed, paste, and mounting have all been checked and the gradual pattern persists, further effort on those layers will not reach the actual restriction. Equally, flushing or refilling a sealed AIO is not a user-serviceable repair: the unit is sealed, the correct coolant and fill procedure are manufacturer-specific, and an incorrect attempt can damage the loop. A cooler with a suspected internal restriction needs workshop assessment, where the loop can be tested as a unit and the cooler replaced if the restriction is confirmed.

Separating Pump and Blockage Problems from Other Overheating

Pump and blockage problems are only part of the overheating picture, and several other causes produce a hot core that must be ruled out before blaming the loop. Radiator fans that have stopped, or a radiator choked with dust, produce a different pattern: the coolant warms up, often substantially, because heat is being picked up but not released, and the radiator feels hot while airflow through it feels weak. A cold plate that has lost contact with the chip — loose mounting, or paste that has dried out and no longer transfers heat — produces a hot core with a loop that stays relatively cool, which resembles a dead pump, but the pump speed reads normally, the tubes may still show flow, and the problem often follows recent work inside the case or a cooler that was removed and refitted. Ambient temperature and case airflow matter too: a loop cannot release heat into air that is already hot or stagnant, so a card that overheats only on hot days or only with the case side panel in a particular configuration deserves an airflow check before any cooler conclusion.

The table below summarises the patterns side by side. No single row should be used on its own; the diagnosis comes from the combination of pump speed, sound, coolant behaviour, tube and radiator feel, and timeline.

A practical diagnostic sequence follows from that table, and it is worth doing in order rather than jumping to the most dramatic conclusion. First, record the baseline: core temperature and coolant temperature, if reported, at idle and under a short, moderate load, plus the pump speed reading and fan speeds. Second, check the simple causes: pump and fan power connections, fan curves and pump control settings, and whether anything changed recently inside the case. Third, listen and feel: pump vibration, gurgling, radiator warmth, and the temperature difference between the two tubes. Fourth, test position: if gurgling or erratic temperatures are present, reposition so the radiator top sits above the pump and see whether symptoms clearly improve. Fifth, review the timeline: sudden and consistent, intermittent and position-sensitive, or gradual over months. Only after those steps does the picture point firmly enough at one of the three loop failures to decide on next steps, and any step that involves opening the loop or removing the cooler from the card should be left to a workshop, because it risks the card itself as well as the cooler.

Common Misconceptions That Send Diagnosis in the Wrong Direction

Several widely repeated beliefs about AIO coolers actively mislead diagnosis. The first is that a loud pump is a strong pump. Noise is not flow. A pump churning air is loud and moves little coolant, while a healthy pump can be nearly silent. Judge flow by temperatures and by the pump speed reading, not by volume. The second is that coolant temperature and core temperature should be close together. They should not be: the core is the heat source and always runs hotter than the coolant, and the gap between them is normal and load-dependent. What matters is how that gap changes — a gap that suddenly becomes very large with a cool loop, or that widens gradually over months, is the signal, not the existence of a gap. The third is that turning radiator fans to maximum compensates for a pump or blockage problem. Fans only help if heat is reaching the radiator; with no flow, or with restricted pickup, faster fans cool a radiator that has little heat to release, and the core stays hot.

A fourth misconception is that a sealed AIO can be topped up like a car radiator. It cannot, in normal user service: there is no user fill port on most sealed units, the coolant is a specific formulation, and opening the loop voids any remaining service coverage and risks leaks onto the card. Coolant loss through permeation over years is a real ageing effect, but its remedy is assessment and replacement of the unit, not a home refill. A fifth is that shaking or tapping the pump clears a clog. Shaking may dislodge a bubble, which is why it sometimes appears to help an airlock, but it does not clear sediment from fine cold plate channels, and vigorous handling while the machine is running risks other damage. Finally, a single monitoring program should never be treated as the whole truth: pump speed, coolant temperature, and fan speed can all be misreported by one tool, and a conclusion as consequential as "the pump is dead" deserves a second tool or a firmware-screen check.

What to Do Next, and When to Seek Inspection

If the pattern points to a dead pump after power and control have been verified, stop load testing and arrange inspection: the cooler is not moving heat, and continued heavy use only accumulates thermal stress. If the pattern points to an airlock, try the radiator-above-pump repositioning once, gently, and observe over several sessions; a lasting improvement confirms the picture, while a quick return of gurgling and erratic temperatures means the loop needs assessment. If the pattern points to a gradual restriction, note the timeline and the widening core-to-coolant gap, and arrange assessment rather than spending further effort on fans, paste, or airflow, which address different layers. In all three cases, keep notes of what was observed — pump speed readings, coolant and core temperatures at idle and under load, sounds, and what changed with position — because those notes let a workshop verify the pattern quickly instead of starting from a bare statement that the card runs hot.

Workshop inspection for cooler and card issues is available at Room 1209, Dali Plaza, Shek Mun, Sha Tin, by telephone on 2186 8010 or WhatsApp on 6558 6806. Published inspection and repair charges follow a three-tier range of HK$400–2,800, and the actual quote for any specific card follows inspection of that card and its cooler, since the fault and the remedy differ from one unit to the next. Bring the whole machine or the card with its cooler attached where possible, so the loop can be assessed as it was running, rather than as separate parts.

Frequently Asked Questions

How can I tell a dead pump from a clogged cold plate?

Check the pump speed reading and the coolant temperature together. A dead pump shows zero or missing pump speed with coolant that barely warms under load. A clogged cold plate shows a normal pump speed with coolant that does warm up, but a core temperature that stays higher than expected for that coolant temperature and worsens gradually over months.

My pump speed reads normally but the card still overheats. Does that rule out the pump?

No. Pump speed reports that the motor is turning, not that coolant is flowing well. An airlock can leave the motor spinning normally while a bubble blocks flow, and a partially blocked cold plate can leave the motor spinning while heat transfer is restricted. Normal pump speed rules out a dead motor only; flow and heat transfer still need the temperature and sound checks.

Is gurgling always a sign of an airlock?

Gurgling at the pump area, especially after startup or after moving the machine, is a strong indicator of air passing through the pump. Brief gurgling that settles quickly and never returns may simply be air redistributing after movement. Persistent gurgling, or gurgling paired with erratic temperatures, points to an airlock that repositioning or professional assessment should address.

Can I fix an airlock by tilting the case?

Gentle repositioning so the radiator top sits above the pump can let air migrate to the radiator, where it causes less trouble, and a clear improvement after repositioning is itself useful diagnostic evidence. Aggressive shaking or repeated tilting under load is not advised, and if symptoms return persistently, the loop needs assessment rather than repeated repositioning.

Why is my coolant temperature low while my core temperature is very high?

That combination means heat is not reaching the coolant effectively. The usual causes are a pump that is not driving flow, air blocking flow at the pump, or poor transfer at the cold plate. Check the pump speed reading first, then listen for gurgling, then check mounting and contact — in that order, because each step rules out a different layer.

Should I open a sealed AIO to refill or flush it?

No. Sealed AIO units are not designed for user refilling or flushing, and opening one risks coolant loss, leaks onto the card, and loss of service coverage. Coolant degradation or internal restriction is assessed as a whole unit in a workshop, where replacement of the cooler is the usual remedy if the restriction is confirmed.

Does a higher fan speed fix a pump failure?

No. Radiator fans release heat that has already reached the radiator. If the pump is not moving coolant, or the cold plate is blocked, little heat reaches the radiator, and faster fans cool a loop that has little heat to give up. Fan speed helps only when flow and pickup are working and the bottleneck is genuinely at heat release.