One GPU Fan Not Spinning or Out of Sync? Single-Fan Failure, Controller & Power Diagnosis Guide

Key Takeaways
When one fan on a multi-fan graphics card stays still or spins at a different speed while the others work under load, the fault needs careful observation, not guesswork. This guide separates normal fan-stop and staged start behaviour from cable, motor, controller and mismatched-part causes.
What This Article Covers, and What It Does Not
This article covers one specific pattern: on a graphics card with two or more fans, one fan behaves differently from its neighbours while the card is under load. It never moves while the others spin, it moves intermittently, it spins slowly, or it spins at a clearly different speed. The goal is to separate the possible layers behind that behaviour: obstruction or blade damage, a fan power or control cable or connector problem, a seized motor or worn bearing, a controller channel or fan-curve profile driving that channel differently, and a mismatched replacement fan.
Several neighbouring topics are marked off first. This article is not about all fans stopped at idle. That pattern has its own explanation and its own guide: many modern cards deliberately stop all fans at low temperature, a feature often called Zero RPM or Fan Stop. This article is not primarily about fan noise, although noise appears below as a supporting clue. Noise on its own belongs to a fan-noise guide. This article is not about temperature faults in general, although temperature readings appear as supporting evidence. And this article is not a repair manual. It is an observation and diagnosis guide that ends at the point where inspection takes over.
Keeping the question narrow matters because a single fan behaving differently is not automatically a fault. Multi-fan cards cool different zones, and depending on design, fans may be driven as a group or on separate channels. Some designs spin the middle fan in the opposite direction to its neighbours. A difference seen at idle, or in the first seconds after start-up, can be entirely normal. The comparison that matters is sustained behaviour under load, against that card's own normal behaviour.
How Multi-Fan Cooling Is Organised
A multi-fan graphics card spreads its fans across different zones of the heatsink. Each fan moves air over a particular region, and the regions do not all heat equally. The processor zone, the memory zone, and the power delivery zone have different thermal behaviour, so a design that treats every fan identically would not always be sensible. That is why fan control varies between designs.
In some designs, all fans are driven as a group: one control signal sets the same behaviour for every fan. In such a design, a single fan behaving differently from its neighbours is more suspicious, because the control input is shared and the difference must come from that fan, its cable, or its mechanical state. In other designs, fans sit on separate channels, and the controller can drive each channel differently. In such a design, a difference can come from the control profile itself, and the diagnosis must check the profile before blaming the fan.
A further design variation is rotation direction. Some cards spin the middle fan in the opposite direction to the outer fans, a design intended to reduce turbulence between neighbouring fans. A middle fan turning the other way is therefore not, by itself, evidence of a fault. Direction must be judged against the card's own design, not against an assumption that all fans should turn the same way.
Fan speed also varies with temperature by design. At idle, many cards stop all fans deliberately. Under light load, fans may start at low speed. Under sustained heavy load, fans should settle into a sustained pattern. Judging a fan from a single glance at idle, or from the first seconds after a load begins, ignores this staged behaviour. The diagnostic window is sustained load, after the start-up phase has settled.
The Key Distinction: Idle Behaviour Versus Under-Load Behaviour
All fans stopped at idle can be normal. That is the Zero RPM or Fan Stop pattern: below a temperature threshold, the controller keeps the fans still because passive cooling is enough. The fans should then start as temperature rises under load. If all fans start together under load and behave consistently, the idle stop was normal behaviour, not a fault.
One fan stopped while the others spin under load is different, and it is this article's subject. The abnormality is the contrast: under the same load, at the same temperature, with the same control conditions, one fan does not match its neighbours. That contrast, sustained over time under load, is what separates a fault pattern from a design pattern.
A fan that starts late also needs care. Some profiles stage fan start-up, bringing fans up in sequence rather than all at once. A fan that starts a few seconds after its neighbours and then settles into matching behaviour is showing a staged start, not a failure. The judgement must rest on sustained load behaviour, not on the first seconds. Only a fan that stays different once load has been sustained counts as a fault pattern.
The same logic applies to speed differences. A small, brief difference during a speed transition is not the same as a sustained, clearly different speed under steady load. Record what each fan does over a sustained period, not in a single snapshot. A snapshot can catch a transition; a sustained observation shows a pattern.
What to Observe Under Load
Observation is the whole method of this guide, and it must be done safely. Load the card with a sustained workload, then watch each fan in turn. Do not poke or flick a spinning fan, and do not reach inside a running machine. The observation below is visual only while the machine is running.
First, identify which fan is different. On a three-fan card, note whether it is the left, middle, or right fan as viewed in the installed position, and write it down. "One fan" is not precise enough for inspection; the position matters, because different positions cool different zones and may sit on different control channels.
Second, describe the behaviour precisely. Does the fan never move at all under sustained load? Does it move intermittently, starting and stopping without matching its neighbours? Does it spin slowly, turning visibly slower than the fans beside it? Or does it spin at a clearly different speed, faster or slower, in a sustained way? Each of these points toward different layers, and vague descriptions blur them together.
Third, note the temperature context. While the fan in question is still or slow, do temperature or hotspot readings climb? Does the card's overall temperature behave differently from what is usual for that workload? A single fan spinning does not prove cooling is adequate, because zone temperatures matter: the zone served by the still fan can overheat while the rest of the card looks acceptable. Conversely, stable temperatures do not prove the fan is fine, because the remaining fans may be compensating for now. Temperature is supporting evidence, not a verdict.
Fourth, note the noise context, as a supporting clue only. A grinding, clicking, or scraping sound from the position of the still or slow fan supports a mechanical layer. A fan that is silent and still, with no mechanical sound, points more toward a control, cable, or motor-electrical layer. Noise alone does not diagnose rotation behaviour, and this guide does not treat it as proof, but it helps inspection know where to look first.
Fifth, check the fan-curve and profile settings. Some control software allows per-fan or grouped control, and a profile can drive one channel differently from the others, whether by design or by an accidental setting change. Record whether the profile uses grouped or per-fan control, and whether any channel has a different curve. A profile difference found here can explain a behaviour difference without any hardware fault. A profile that treats all fans identically, combined with one fan still behaving differently, pushes the diagnosis toward the fan, its cable, or its channel.
Sixth, compare with idle behaviour. Confirm the idle pattern separately: at idle, do all fans stop together, or does the same fan behave differently even at idle? A fan that is different at idle and under load, and a fan that is normal at idle but different under load, carry different implications, and the comparison helps inspection narrow the layer. Write both observations down.
The Possible Layers, in Order
The first layer is obstruction or blade damage. Something blocking the fan, or a damaged blade, prevents normal rotation. This layer is checked only with the machine off and power disconnected: a visual inspection for an obstruction, a cable resting against the blades, or visible blade damage. Never inspect for obstruction while the machine is running, and never flick a blade to test it.
The second layer is the fan power or control cable, or its connector. A cable that is not seated, or a connector that has worked loose, can cut power or control to one fan while the others run normally. This layer is difficult to confirm from outside, because the cable and connector sit within the card assembly. The observation that points here is a fan that is entirely still, with no mechanical sound and no intermittent movement, while its neighbours behave normally and the profile treats all fans identically. That pattern is consistent with a cable or connector problem, but it is also consistent with a motor problem, and the two cannot be separated reliably from outside.
The third layer is the fan motor or bearing. A seized motor or worn bearing prevents rotation or makes it slow and uneven. Intermittent movement, slow spinning, wobbling, or a grinding sound point more toward this layer. A fan that tries to move but cannot sustain rotation, or that spins unevenly, is showing mechanical difficulty rather than a clean absence of power. A visibly tilted or wobbling fan is a mechanical problem, not a setting, and it belongs in the stop points below.
The fourth layer is the controller channel or fan-curve profile. If the profile drives channels differently, or if one channel is faulty, a single fan can behave differently while the fan itself is healthy. The observation that points here is a behaviour difference that follows the profile: changing the profile changes the difference, or the difference matches a per-channel setting. If no profile difference exists and the fan still differs, this layer becomes less likely, but a faulty channel cannot be ruled out from outside.
The fifth layer is a mismatched replacement fan. If a fan has been replaced, a part that does not match the original can spin at a different speed under the same control signal, because its motor characteristics differ. The observation that points here is a speed difference that began after a fan replacement, in a fan that otherwise rotates smoothly and without mechanical distress. A fan that spins but at a different speed after replacement is a compatibility clue, not a control mystery.
The following table summarises how observations map to layers. It is a guide to narrowing the range, not a proof of any single layer. Several layers produce overlapping symptoms, and the final distinction requires inspection.
| Observation under sustained load | Most likely layer | Supporting check |
|---|---|---|
| Fan never moves, no mechanical sound, neighbours normal, profile treats all fans alike | Cable or connector, or motor electrical failure | Confirm profile is grouped or identical across channels; record which fan and that movement is entirely absent |
| Fan moves intermittently, or spins slowly and unevenly | Motor or bearing difficulty | Note any grinding or clicking from that fan position; do not force the fan by hand |
| Fan wobbles or sits visibly tilted | Mechanical problem | Stop and seek inspection; do not run sustained load |
| Fan spins smoothly but at a clearly different sustained speed, after a fan replacement | Mismatched replacement part | Record when the replacement happened and that rotation is smooth but mismatched |
| Behaviour difference follows a profile change, or matches a per-channel setting | Controller channel or fan-curve profile | Record the profile settings before and after the change |
| Fan starts late but then matches its neighbours under sustained load | Staged start-up profile, normal behaviour | Judge by sustained behaviour, not the first seconds |
| All fans stopped at idle, all start together under load | Zero RPM or Fan Stop, normal behaviour | Confirm the under-load pattern is consistent across fans |
| Temperatures climb while one fan stays still under sustained load | Cooling for that zone is compromised, whatever the layer | Do not run sustained heavy load in this state; record readings with the load state |
Safe Handling Rules
Several handling rules protect both the observer and the evidence. Never poke or flick a spinning fan. Never force a seized fan to spin by hand while power is connected. Only inspect for obstruction with the machine off and power disconnected, and even then, do not dismantle the card. Do not run sustained heavy load with a dead fan while temperatures keep climbing: a short observation under load is diagnostic, but prolonged load in that state risks damage to the zone that has lost its cooling. Do not alter fan wiring, swap fan cables between channels, or fit an unmatched fan as a test. Those actions belong to inspection, where the card can be handled properly and the replacement part can be matched.
These rules also protect the diagnosis. Forcing a fan to move, or changing cables, alters the very behaviour being observed. A fan that was forced to spin once may behave differently afterwards, and the record of its natural behaviour is lost. Observation first, untouched, is worth more than any improvised intervention.
Edge Cases That Are Easy to Misread
The first edge case is the late-starting fan. A fan that starts after its neighbours can be a staged start-up profile, not a failure. Judge by sustained load behaviour. If the late fan settles into matching behaviour and stays matched, the start delay was a profile characteristic. If it stays different, the delay was the beginning of a fault pattern.
The second edge case is opposite rotation direction. Some designs spin the middle fan in the opposite direction to its neighbours. A middle fan turning the other way, at a matching speed and with normal cooling, is a design characteristic, not a fault. Direction alone, without a speed or stillness difference, is not diagnostic.
The third edge case is a fan that spins after a replacement but at a different speed. This points to a mismatched part rather than a new fault in the controller or the remaining fans. The diagnostic step is to record the replacement history, not to adjust the profile to force a match. Forcing a mismatched fan to match by profile can mask the mismatch while leaving the underlying incompatibility in place.
The fourth edge case is noise without rotation difference. A noisy fan that still rotates at a matching speed is a noise problem, not a rotation problem. It belongs to the fan-noise guide. Noise becomes relevant to this guide only when it accompanies a rotation difference, as a supporting clue pointing toward the mechanical layer.
The fifth edge case is a fan that looks still because it is viewed at the wrong moment. A fan at low speed can be hard to judge in a brief glance, and a fan in transition can appear still in a snapshot. Sustained observation, watching each fan in turn over a period under load, avoids this error. If there is doubt, observe again under the same load rather than concluding from one glance.
The sixth edge case is a difference that appears only at one particular load level. A fan that matches its neighbours at moderate load but differs at maximum load, or the reverse, is still showing a fault pattern, but the load dependence is itself informative. Record the load level at which the difference appears and disappears. A difference tied to a load threshold points inspection toward the control or motor behaviour at that operating point.
Common Misconceptions
Misconception one: one stopped fan means the whole cooler has failed. It does not. The cooler is a multi-fan system, and the remaining fans continue to move air. The risk is zone-specific: the region served by the still fan loses its cooling, even while the rest of the card is cooled. The correct conclusion is narrower and more serious than a total failure claim: one zone is compromised, and sustained heavy load in that state is unsafe.
Misconception two: setting fans to maximum diagnoses a dead channel. It does not. A maximum setting changes the control input for all channels, but a fan with a cable, motor, or channel problem will still not respond normally. The test adds noise and heat without separating layers. The diagnostic comparison is between fans under the same normal profile, not under an extreme override.
Misconception three: a single fan spinning proves cooling is adequate. It does not. Zone temperatures matter. The spinning fans cool their own zones; the zone of the still fan can still overheat. Adequacy must be judged by temperature readings across zones under sustained load, together with fan behaviour, not by the sight of one fan moving.
Misconception four: "it still works" means sustained load is safe. It does not. A card can continue to function while one zone overheats gradually, with damage accumulating silently. Continued function under light load says nothing about safety under sustained heavy load with a dead fan. The stop points below exist precisely because continued operation can look deceptively normal.
Misconception five: a speed difference must be a setting problem. Settings are one layer, but motor, bearing, cable, and mismatched-part layers all produce speed differences too. A speed difference that survives a profile check, and that began without any settings change, points away from settings. Checking the profile is necessary, but it is not the whole diagnosis.
Misconception six: all fans should behave identically at all times. Design differences in channel control, staged start-up, and rotation direction mean identical behaviour is not the baseline. The baseline is that card's own normal behaviour under sustained load. Judging a fan against a generic expectation, rather than against its neighbours under the same load and its own design, produces false alarms.
Stop Points: When Not to Test Further
Stop observation and seek inspection if any of the following occur. First, the fan is detached, wobbling, visibly tilted, or grinding. These are mechanical problems, not settings. Second, any unusual smell, scorch marks, or abnormal heat from the machine. Third, temperatures climbing under load while the fan stays still, especially if the climb continues rather than stabilising. Fourth, any urge to dismantle the card, alter fan wiring, swap cables, or fit an unmatched fan. Fifth, observation is complete and points to a dead channel, but the layer cannot be narrowed between cable, motor, and controller from outside. That uncertainty is itself a stop point: further testing from outside will not resolve it, and inspection is the next step.
Stopping is not giving up. A precise record, stating which fan, what it did under sustained load, what the temperatures did, and what the profile showed, is the most useful material inspection can receive. Continuing to run heavy load with a dead fan, or intervening by hand, only risks damage and blurs that record.
Information to Record Before Seeking Inspection
Record which fan is affected, by position, and exactly how it behaves under sustained load: never moves, moves intermittently, spins slowly, or spins at a clearly different speed. Record the idle pattern separately, including whether all fans stop together at idle. Record temperature or hotspot readings during the observation, together with the load state at the time, never as a lone figure. Record the fan-curve or profile settings, including whether control is grouped or per-fan, and whether any channel differs. Record any fan replacement history, any settings change before the problem appeared, and any mechanical sound from the affected fan's position. Describe facts only, without inferring the layer yourself. A written record lets inspection start at the narrowed range instead of repeating the whole observation sequence.
Frequently Asked Questions
Q: All my fans stop at idle. Is one of them faulty?
A: Not on that evidence alone. Many cards deliberately stop all fans at low temperature, a feature often called Zero RPM or Fan Stop. The test is under load: if all fans start together and behave consistently under sustained load, the idle stop was normal behaviour.
Q: One fan stays still while the others spin under load. Is that normal?
A: No, that contrast under sustained load is abnormal, and it is the subject of this guide. Normal design differences, such as staged start-up or opposite rotation direction, do not leave one fan still while its neighbours spin under sustained load.
Q: My middle fan spins the opposite way. Is it broken?
A: Not necessarily. Some designs spin the middle fan in the opposite direction to reduce turbulence between neighbouring fans. Direction alone is not a fault. Judge by whether the fan rotates at a matching speed and whether cooling stays normal under sustained load.
Q: One fan starts a few seconds after the others. Should I worry?
A: Judge by sustained behaviour, not the first seconds. Some profiles stage fan start-up. If the late fan settles into matching behaviour under sustained load, the delay was a profile characteristic. If it stays different, the delay was the start of a fault pattern.
Q: Setting fans to maximum did not fix the still fan. What does that tell me?
A: It tells you the problem is not a low-speed profile setting, but it does not separate the remaining layers. A cable, motor, or channel problem will not respond to a maximum setting either. The useful comparison is between fans under the same normal profile, recorded precisely.
Q: The card still works with one fan stopped. Can I keep using it heavily?
A: No, sustained heavy load is not safe in that state. The remaining fans cool their own zones, but the zone served by the still fan can overheat. A short observation under load is diagnostic; prolonged heavy load while temperatures climb risks damage.
Q: A fan was replaced and now spins at a different speed. Is the controller faulty?
A: More likely the replacement part is mismatched. A fan with different motor characteristics can spin at a different speed under the same control signal while rotating smoothly. Record the replacement history and seek inspection for a matched part, rather than forcing a match through profile settings.
Q: Can I tell from outside whether it is the cable, the motor, or the controller?
A: Usually not reliably. Observation narrows the range: a silent, entirely still fan with an identical profile points toward cable or motor-electrical layers, while intermittent or uneven movement points more toward motor or bearing difficulty, and a profile-linked difference points toward control. The final distinction between cable, motor, and controller requires inspection.
Conclusion: Compare Fans Under Sustained Load, Then Stop at the Boundary
Single-fan diagnosis is an exercise in disciplined comparison. Watch each fan under sustained load, compare it with its neighbours under the same conditions, check the profile, and record temperatures and positions precisely. Normal design differences, staged starts, and idle fan-stop must be excluded first, so that a genuine fault pattern is not confused with a design characteristic. Once observation points to a dead channel without certainty between cable, motor, and controller, or once any mechanical warning sign appears, the observation phase is complete and inspection takes over.
If that point is reached, or if temperatures climb while a fan stays still, bring the written observation record when seeking inspection. That record is far more useful than saying one fan does not spin. HK GPU Fix is located at Room 1209, Dali Plaza, Shek Mun. Phone 2186 8010, WhatsApp 6558 6806.