GPU Fan Keeps Starting and Stopping Near One Temperature? Start-Stop Threshold, Hysteresis, Curve Steps & Sensor Fluctuation Diagnosis Guide

2026-11-04 | GPU Repair | 17 min | HK GPU Fix Editorial

GPU Fan Keeps Starting and Stopping Near One Temperature? Start-Stop Threshold, Hysteresis, Curve Steps & Sensor Fluctuation Diagnosis Guide

Key Takeaways

When a graphics card fan repeatedly starts and stops while the temperature hovers near a single point under light, steady load, the cycling itself is the clue. This guide explains the gap between switch-on and switch-off points, stepped fan curves, sensor wobble, and control-source confusion, and shows how to tell which one explains the pattern.

The Decisive Feature: Cycling Near One Point, Under Light and Steady Load

This guide is about one narrow pattern, and the pattern has to be stated exactly before any explanation is attempted, because fan behaviour has many normal forms that look similar at a glance. The fan repeatedly starts and stops in a cycle, switching on, then off, then on again, while the temperature hovers near one threshold, and while the load is light and steady. The cycling is the symptom. The hovering near one point is the context that gives the cycling its meaning. The light, steady load is the condition that rules out load swings as the explanation.

Three nearby behaviours must be separated from this pattern at the start, because confusing them with cycling leads to false conclusions. Under sustained heavy load, a fan that stays on is not cycling. It is doing its job continuously, because the heat being produced requires continuous cooling. At true idle, when the card is cool, a fan that stays off is not cycling either. A steady-off state at low temperature, sometimes described as a zero-speed or stopped-at-idle state, is a stable state, not a cycle, and it should not be treated as a fault simply because the fan is not turning. Cycling is neither of those stable states. It is repeated switching between on and off while conditions stay roughly the same, and that repetition under stable conditions is what needs explaining.

The load condition matters as much as the temperature condition. Under a load that rises and falls sharply, a fan that follows the load will naturally switch more often, and that following behaviour is a different story from the one told here. Under light, steady load, the heat being produced is roughly constant, so a fan that keeps switching on and off is not simply following a changing load. Something in the control logic, the curve shape, the sensor reading, or the temperature source being followed must be producing the switching, and the rest of this guide separates those four explanations.

Observation comes before any change, and it should be patient. Watch whether the fan switches on and off in a repeating rhythm, or irregularly. Watch whether the temperature reading rises and falls in step with the fan switching, or stays roughly steady while the fan switches anyway. Watch whether the cycling continues for a long stretch under the same light load, or appears only briefly during a transition between states. Those observations, recorded over a stretch of steady load rather than judged from a single glance, are the evidence that later separates the four mechanisms. A single moment of fan noise proves nothing. A pattern observed over time proves a great deal.

Mechanism One: Hysteresis, or Why Switch-On and Switch-Off Are Not the Same Point

The first mechanism is the most important, and it is best understood in plain language before any technical term is attached to it. Imagine a control that switches a fan on when the temperature rises past a point, and switches it off again when the temperature falls past a point. If those two points are exactly the same, the control has a problem. The moment the temperature crosses the point upward, the fan starts. The fan cools the card slightly, the temperature falls back across the same point, and the fan stops. Without cooling, the temperature rises across the point again, and the fan starts again. The result is rapid, repeated switching, known as chatter, driven not by any fault in the fan, but by the logic of using a single point for both decisions.

The remedy built into well-designed controls is a gap between the two points. The fan switches on at a higher point and switches off at a lower point, so that after starting, the temperature must fall a meaningful distance before the fan stops, and after stopping, it must rise a meaningful distance before the fan starts again. That gap is called hysteresis, and its purpose is exactly to prevent chatter. It turns a knife-edge decision into a band with room on either side, so that small temperature wobbles around a point do not produce switching.

Cycling near one threshold is what happens when that gap is narrow or missing. With a narrow gap, a small rise starts the fan, a small fall stops it, and the cycle repeats, because the temperature never has to travel far in either direction to cross a switching point. With no effective gap, the behaviour approaches the single-point chatter described above. The key diagnostic idea follows directly: hysteresis problems produce cycling that tracks small temperature movements around the switching band. The fan starts as the reading crosses upward, stops as it crosses downward, and the cycle repeats as the reading wobbles across the band. That tracking relationship, fan switching in step with small temperature crossings, is the signature of a gap problem, and it is distinct from the signatures of the other mechanisms below.

It is important to state what this section does not claim. It gives no numerical values for where any switch-on or switch-off point should sit, because those values vary by design and by card, and inventing a figure would mislead. The mechanism is described in terms of relationships, a higher switch-on point, a lower switch-off point, and a gap between them, and the diagnosis in this guide works from observed relationships, not from invented numbers. A reader who wants to know the actual points for a particular card should look to that card's own documentation, not to a general guide.

Mechanism Two: A Stepped Fan Curve, Where One Step Is Crossed Again and Again

The second mechanism concerns the shape of the fan curve, meaning the relationship between temperature and fan speed as set in the control. Some curves rise smoothly, increasing fan speed gradually as temperature rises. Others rise in steps, holding one speed over a range of temperatures and then jumping to a higher speed at a particular temperature. A stepped curve is not inherently wrong, but a step creates a sharp boundary, and a sharp boundary interacts badly with small temperature wobbles.

Picture a step placed at one temperature. Below the step, the fan is off or slow. Above the step, the fan is on or faster. If the temperature hovers near the step, drifting slightly above it and slightly below it, the control crosses the step repeatedly, and the fan switches each time it crosses. The fan is not faulty, and the temperature is not wildly unstable. The combination of a sharp step and a hovering temperature is enough to produce cycling, because every small wobble becomes a crossing, and every crossing becomes a switch.

The diagnostic signature of a step problem is similar to that of a hysteresis problem, in that cycling tracks temperature crossings, but the emphasis is different. In a hysteresis problem, the issue is the gap between two switching points being too small. In a step problem, the issue is the shape of the curve itself, a jump rather than a slope at the temperature where the card happens to be hovering. In practice the two can combine: a stepped curve with a narrow gap around its step is especially prone to cycling, because the step provides the sharp boundary and the narrow gap provides no buffer against wobble. Separating them conceptually still matters, because the setting-level discussion that follows differs: a gap problem invites discussion of widening the gap, while a step problem invites discussion of smoothing the step, meaning replacing a jump with a gradual slope so that small wobbles produce small speed changes rather than on-off switching.

That setting discussion must be kept at the right level. Smoothing a step or widening a gap is a matter of fan control settings, and this guide discusses it only as a setting-level direction, not as a procedure. It gives no step-by-step instructions for editing driver settings or system settings, and it names no brand menu paths, because a general guide cannot safely prescribe a procedure that varies by software and by card, and because a reader who changes settings without understanding the control risks trading one problem for another. The proper use of this section is diagnostic: to recognise that a step at the hovering temperature is a candidate explanation, and to carry that recognition into a careful, documented setting discussion, rather than into hasty changes.

Mechanism Three: Sensor Reading Fluctuation Crossing the Threshold

The third mechanism does not require the true temperature to wobble at all. It requires only that the reading wobbles. A temperature sensor reports a value, and that reported value can fluctuate slightly even when the underlying temperature is roughly steady, because sensing and reporting involve measurement, and measurement always carries some variation. If the reported value hovers near a switching point, its small fluctuations can cross that point repeatedly, and the control, which acts on the reported value rather than on some hidden true value, switches the fan each time it crosses.

The diagnostic signature here is the opposite, in an important respect, from the first two mechanisms. In hysteresis and step problems, the cycling tracks genuine small temperature movements around a band or a step. In a sensor fluctuation problem, the cycling happens even when the temperature, as far as can be judged, is steady. The fan switches while the broader thermal picture stays flat, because the reading itself is doing the crossing. That distinction is subtle, and it cannot be judged from a single reading. It requires watching the reported temperature over a stretch of steady load and noticing whether the cycling aligns with visible rises and falls in the reading, or proceeds while the reading stays within a narrow, flat range apart from tiny flickers at the switching point.

Sensor fluctuation also carries a caution against over-interpretation. A reading that flickers slightly is not, on its own, proof of a faulty sensor. Small variation in a reported value is a normal feature of measurement, and it only becomes a cycling problem when it interacts with a switching point placed where the reading hovers. The fault, if that word is even appropriate, lies in the interaction between the reading, the switching point, and the gap, not necessarily in any single component. That is why this guide treats sensor fluctuation as one candidate among four, to be weighed against the others from observation, rather than as a verdict to be jumped to.

Mechanism Four: Fan-Control Source Confusion, or Which Temperature the Curve Follows

The fourth mechanism is the easiest to overlook, because it hides in an assumption: the assumption that the temperature being watched is the temperature the fan curve follows. A graphics card has more than one temperature that can be sensed, and a fan curve follows one particular source. If the person watching the behaviour is looking at a different source from the one the curve follows, the fan behaviour can look mysterious. The fan may switch while the watched temperature looks steady, or stay steady while the watched temperature moves, simply because the curve is responding to a different measurement from the one on display.

Source confusion produces a distinctive kind of puzzlement. The cycling does not track the watched temperature in any clean way, because the watched temperature is not the controlling one. The fan may appear to switch for no reason, or to ignore a temperature change that seems obvious, and the observer may conclude that the fan or the control is faulty, when in fact the control is behaving consistently with respect to its own source. The remedy is not a part replacement. It is identifying which temperature source the curve follows, and then judging the fan behaviour against that source, rather than against a different reading that happens to be on screen.

This mechanism also explains why comparing notes between different setups can mislead. Two cards that appear similar may follow different sources, or may display different sources by default, so that the same fan behaviour means different things in each case. The diagnostic habit that protects against this confusion is to state, for any observation, which temperature is being watched, and to treat that statement as part of the evidence. An observation that does not say which source it refers to is incomplete, in the same way that a cable test that does not say which end was reseated is incomplete.

The Diagnostic Habit: Does Cycling Track Wobble, or Happen at Steady Temperature?

The four mechanisms converge on one diagnostic habit, and that habit is the core practical skill of this guide. Under light, steady load, watch the relationship between fan switching and temperature over a stretch of time, and ask one question: does the cycling track temperature wobble, or does it happen while the temperature stays steady?

If the fan starts as the reading rises across a point and stops as it falls back across it, repeating in step with small rises and falls, the evidence leans toward a gap problem, a step problem, or a combination of the two, because both produce switching that tracks crossings. The next question then separates gap from step: is the switching organised around a band with two points, or around a single sharp jump in speed? That question is answered by observing the shape of the behaviour, whether the fan moves between off and on across a band, or jumps at one sharp boundary, and by considering the curve shape as a setting-level discussion.

If the fan switches while the temperature picture stays broadly flat, the evidence leans toward sensor fluctuation or source confusion, because both produce switching that does not track the watched temperature cleanly. The next question then separates those two: does the reported value itself flicker across the switching point, which leans toward fluctuation, or does the fan behave consistently once the controlling source is identified, which leans toward source confusion? That question is answered by watching the reported value closely and by establishing which source the curve follows, before any setting is changed.

In every branch, change one thing at a time. If a setting-level discussion leads to a change, such as smoothing a step or widening a gap, make that single change, return to the same light, steady load, and observe whether the cycling changes. Changing a setting and also changing the load, the ventilation, or the displayed temperature source in the same step destroys the information the change was supposed to provide. The same discipline that governs cable testing governs fan diagnosis: one variable at a time, with the result noted after each step.

Observed pattern under light, steady loadMechanism it leans towardWhat to observe or discuss next
Fan switches in step with small temperature rises and falls across a bandNarrow or missing hysteresis gapWhether widening the gap, as a setting-level discussion, changes the cycling
Fan jumps on and off at one sharp speed boundary where temperature hoversStepped fan curveWhether smoothing the step into a gradual slope, as a setting-level discussion, changes the cycling
Fan switches while the broader temperature picture stays flat, with the reading flickering at the pointSensor reading fluctuation crossing the thresholdWhether the reported value itself flickers across the switching point over a stretch of steady load
Fan switches with no clean relationship to the temperature on displayFan-control source confusionWhich temperature source the curve actually follows, and whether behaviour is consistent against that source
Fan stays on under sustained heavy loadNormal continuous cooling, not cyclingNo cycling diagnosis applies; heavy load requires continuous cooling
Fan stays off at true idle while coolSteady-off state, not cyclingNo cycling diagnosis applies; a stable off state at low temperature is not a cycle
Fan switches only during load transitions, not under steady loadLoad-following behaviour, not the pattern in this guideObserve under a longer stretch of steady load before applying this guide
Cycling persists unchanged after one clean setting discussion changeMore than one mechanism, or a different causeRecord every step and result, and seek help with that record rather than stacking further changes

Boundary Cases: When This Guide Does Not Apply

Boundaries protect readers from applying a good explanation to the wrong behaviour, and this guide has several. A fan that stays off steadily at idle, while the card is cool, is a steady-off state, not cycling, and it needs no cycling diagnosis. A fan that runs at full speed during boot, before control takes over, is a start-up behaviour, not the light-load cycling described here. This guide is not a curve-setting guide: it explains mechanisms and diagnostic habits, and it discusses smoothing and gap widening only as setting-level directions, without prescribing procedures. A single fan that is fully dead, never turning under any load, is a different problem from a fan that repeatedly starts and stops, because a dead fan shows no switching at all. An anomaly that appears only after waking from sleep, and does not recur under steady load, belongs to a wake-related branch, not to the steady-load cycling pattern. Behaviour that appears after gaming, where a clock or fan state seems stuck rather than cycling, is likewise a different branch, because stuck is not cycling. Keeping these boundaries explicit stops a reader with a different behaviour from spending time on threshold explanations that cannot help, and from drawing a false conclusion about a fan that was never faulty in the cycling sense.

Common Misconceptions

Misconception one: a cycling fan is necessarily a faulty fan. Wrong. Cycling under light, steady load near one threshold is often a control, curve, reading, or source story, not a mechanical fan failure. A fan that starts, stops, and starts again on command is demonstrating that it can turn, which points away from a dead fan and toward the logic that commands it.

Misconception two: cycling means the card is overheating. Wrong. Cycling near one threshold under light load is a switching story, not a temperature verdict. Overheating would be judged from the temperature picture as a whole, including sustained load behaviour, not from the mere fact that a fan switches.

Misconception three: a steady-off fan at idle is the same as a cycling fan. Wrong. Steady-off is a stable state at low temperature. Cycling is repeated switching under roughly stable conditions. Treating a stable state as a cycle leads to interventions that fix nothing, because there is no cycle to fix.

Misconception four: changing a setting quickly will settle which mechanism is responsible. Wrong, unless the change is made cleanly, one at a time, and observed under the same steady load. A setting change combined with a load change, a ventilation change, or a change of displayed temperature source proves nothing about which mechanism was responsible.

Misconception five: the temperature on screen is always the temperature the fan follows. Wrong. A curve follows one particular source, and the displayed reading may be a different source. Judging fan behaviour against the wrong source makes consistent control look erratic, and it is one of the most common sources of confusion in this topic.

Misconception six: a single reading settles the question. Wrong. Cycling is a pattern over time, and its mechanisms are distinguished by relationships observed over a stretch of steady load, whether switching tracks wobble or proceeds at steady temperature. A single moment of fan noise, or a single temperature value, cannot distinguish the four mechanisms.

What Not to Assume, and When to Stop

Do not assume a fan is faulty because it cycles, do not assume a temperature figure that this guide does not give, and do not follow step-by-step setting procedures from a general guide that cannot know the software or card in front of you. Setting discussions in this guide stay at the level of direction, smoothing a step or widening a gap, and any actual change should be made carefully, one change at a time, with the result observed under the same steady load. Stacking several setting changes together, or changing settings while also changing load or ventilation, destroys the diagnostic information.

Stop further changes and seek help if any of the following appear: a fan makes grinding or scraping noises, a fan blade is visibly damaged, the card shows other abnormal behaviour beyond the cycling pattern, or a clean sequence of observations and one-at-a-time setting discussions has been worked through and the cycling persists without any branch gaining clear support. Those situations go beyond threshold diagnosis, and they should be handed to inspectors to investigate from a wider angle.

When seeking help, bring records sufficient for the inspector to start directly from the located side: that the cycling occurs under light, steady load near one threshold, whether the switching tracks temperature wobble or happens at steady temperature, which temperature source was watched, whether sustained heavy load keeps the fan on and true idle keeps it steadily off, and which single setting discussion change if any was tried and what it changed. Stated clearly, these are enough for the other party to judge the next step quickly.

If you live nearby, you are welcome to bring your records to HK GPU Fix for advice, at Shatin Dali Plaza Room 1209, phone 2186 8010, WhatsApp 6558 6806. When asking, explain the cycling pattern, whether it tracks temperature wobble or happens at steady temperature, and which temperature source was watched, and that is enough for the other party to judge the next step quickly.

Frequently Asked Questions

Is a fan that keeps starting and stopping always broken?

No. Under light, steady load near one threshold, repeated starting and stopping is often a control story: a narrow or missing gap between switch-on and switch-off points, a sharp step in the fan curve, a sensor reading flickering across a threshold, or a curve following a different temperature source from the one on display. A fan that starts and stops on command is demonstrating that it can turn, which points toward the control logic rather than toward a dead fan.

What is hysteresis in plain language?

It is the gap between the point where a fan switches on and the point where it switches off. The fan switches on at a higher point and off at a lower point, so small temperature wobbles do not cause switching. If that gap is narrow or missing, small wobbles cross a switching point repeatedly, and the fan cycles. The gap exists to prevent chatter, and a missing gap causes it.

Why does a stepped fan curve cause cycling?

Because a step is a sharp jump in speed at one temperature. If the temperature hovers near that step, drifting slightly above and below it, the control crosses the step repeatedly, and the fan switches each time. A gradual slope would turn the same small wobbles into small speed changes instead of on-off switching, which is why smoothing a step is discussed as a setting-level direction.

Can sensor fluctuation cause cycling even when the temperature is steady?

Yes. The control acts on the reported value, and if that reported value flickers across a switching point while the broader temperature picture stays flat, the fan will switch each time the reading crosses. The signature is cycling that proceeds while the temperature picture stays broadly steady, apart from tiny flickers at the switching point.

Why does the fan sometimes seem to ignore the temperature on my screen?

Because the fan curve may follow a different temperature source from the one on display. A card has more than one temperature that can be sensed, and judging fan behaviour against a source the curve does not follow makes consistent control look erratic. Identifying which source the curve follows, and judging behaviour against that source, resolves much of this apparent mystery.

How do I tell whether cycling tracks temperature wobble or happens at steady temperature?

Watch the relationship over a stretch of light, steady load, not from a single glance. If the fan starts as the reading rises across a point and stops as it falls back, repeating in step with small rises and falls, the cycling tracks wobble. If the fan switches while the broader picture stays flat, the cycling happens at steady temperature, which leans toward sensor fluctuation or source confusion.

Is a fan that stays off at idle a problem?

Not on that fact alone. At true idle, while the card is cool, a fan that stays off steadily is in a stable off state, not a cycle. Cycling means repeated switching under roughly stable conditions. A stable off state at low temperature should not be treated as a fault simply because the fan is not turning.

Is a fan that stays on under heavy load cycling?

No. Under sustained heavy load, a fan that stays on continuously is providing continuous cooling for continuous heat, which is normal. Cycling is defined by repeated switching under light, steady load near one threshold, not by continuous operation under heavy load.

Should I just change fan settings until the cycling stops?

No, not as a blind search. Make one setting-level change at a time, such as a discussion of smoothing a step or widening a gap, return to the same light, steady load, and observe whether the cycling changes. Stacking changes together, or changing settings while also changing load or ventilation, destroys the information needed to know which mechanism was responsible. This guide discusses settings only at that directional level, without step-by-step procedures.

When should I stop changing settings myself?

Stop if a fan makes grinding or scraping noises, if a blade is visibly damaged, if the card shows other abnormal behaviour beyond the cycling pattern, or if a clean sequence of observations and one-at-a-time setting discussions has been worked through and the cycling persists without any branch gaining clear support. Bring the cycling pattern, the wobble-versus-steady observation, and the temperature source watched, and hand the problem to inspectors to investigate from a wider angle.

Related articles

Related services

If you need Cooling repair, see these service pages: