GPU Fan Keeps Starting and Stopping Near One Temperature? Start/Stop Thresholds, Fan Curves & Sensor Location Diagnosis Guide

Key Takeaways
When a GPU fan cycles on and off near one temperature, the fan is not necessarily faulty: the key question is whether the gap between the start and stop thresholds is wide enough, and which sensor is driving the curve in practice.
What Exactly Counts as This Symptom
This guide deals with a narrowly shaped symptom, and the shape has to be matched first. The graphics card fan is neither stopped for good nor spinning fast for good. Instead, it repeats the same cycle inside a narrow temperature band: spin for a while, stop for a while, spin again, stop again. Each start may come with a faint spin-up sound, and each stop brings the case back to quiet. The cycle can appear under light loads such as office work, video playback, or a browser session, with no gaming required. The defining feature is rhythm: on-off alternation, with each event happening at roughly the same temperature, rather than random jumping around.
Several lookalike situations must be ruled out first, or the whole diagnosis will head in the wrong direction. First, a fan that stays stopped under light load while temperatures remain low is the silent design working as intended, not a fault. Second, one fan stopped while the others keep spinning is a single-fan issue, not a whole set cycling together. Third, a fan that is plainly spinning while monitoring software reports zero RPM is a reading problem, not a cycling problem. Fourth, a fan that is noisy all the time at a steady speed is a noise problem, not a cycling-rhythm problem. Match the symptom before changing any setting, because a fan curve change aimed at the wrong symptom will not fix anything.
The Mechanism: How One Temperature Can Trap a Whole Fan Set
To understand the cycle, start with the purpose of the fan-stop design. Many graphics cards stop their fans completely at low temperatures and low loads, keeping the case quiet and avoiding needless wear. When temperature rises to a start threshold, the fans begin to spin; when temperature falls back, they stop again. The design itself is sound. The trouble begins when the gap between the start threshold and the stop threshold is too narrow. With almost the same temperature for starting and stopping, any small temperature wobble crosses a threshold. Light loads are exactly the kind of workload that produces such wobbles: opening a web page, decoding a video, or a background task nudges temperature up slightly; the fan spins, temperature drops quickly; the fan stops, temperature creeps back up. The cycle then locks itself into the same temperature band, with no component damage required at all.
The second mechanism is hysteresis. Hysteresis means the on point and the off point are two different temperatures, with a gap left between them. A fan might start at a higher temperature but only stop once temperature has fallen to a lower one, and that in-between gap is the buffer. A wide buffer means small wobbles never trigger a switch; a narrow buffer, or no buffer at all, means the fan reacts to every small wobble. Some software achieves the same effect differently, using a sliding average or a delayed response: temperature must stay above or below a threshold for a stretch of time before the fan changes state, so brief spikes are ignored. The two approaches differ in method but share the same goal, which is to stop the fan from being oversensitive to short-lived fluctuations. This is why the core diagnostic question is not "can the fan spin" but "is the fan oversensitive to temperature change".
The third mechanism is the shape of the fan curve itself. A fan curve is a lookup table: a given temperature maps to a given speed. If the curve is very steep near the threshold, a small temperature difference produces a large speed difference, and the fan response looks abrupt. If the curve sits at zero speed across a long low-temperature stretch, a light-load temperature hovering in that stretch will keep jumping between zero and spin-up. By contrast, a curve that already holds a low, steady speed at low temperatures never needs to stop, so the cycle disappears. There is a trade-off here: a steady low speed carries a little background noise but buys stability, while a full stop is quietest but most prone to cycling near the threshold. No choice is best in every respect; the question is which cost the user is willing to accept.
| Observation | More likely explanation | What to do next |
|---|---|---|
| Fans cycle on and off rhythmically in one temperature band | Start and stop thresholds are too close, so small wobbles keep crossing the line | Log temperature against on/off events, then test a wider gap |
| Fans stay stopped, temperatures stay low, no cycling | Silent design working normally | No action needed; check that temperatures still rise normally under load |
| Fans still cycle irregularly at a fixed speed | Curve and thresholds are not the main cause; control or power delivery is suspect | Stop adjusting curves and record the result for inspection |
| Only one fan stops while the rest keep spinning | Single-fan issue, not a whole-set threshold issue | Do not apply this guide's curve method; diagnose the single fan separately |
| Temperature keeps rising and fans never spin | Cooling control is not responding at all, a different and more urgent situation | Reduce load immediately and seek inspection |
Step One: Log Temperature Against On/Off Events
Before changing any setting, make one clean record. Open monitoring software that shows graphics card temperature and fan speed together, and watch under one fixed light-load scenario, noting three things: the temperature when the fan starts, the temperature when it stops, and how long passes between events. Together, those three items show how narrow the threshold gap is. If the start and stop temperatures are close together and the cycle is frequent, a too-narrow gap is strongly suspected. If the start and stop temperatures are actually well apart yet the fan still cycles frequently, then either the temperature reading itself is swinging wildly, or the sensor driving the fan is not the one being watched.
A few details matter while logging. First, keep the scenario fixed: the same application and the same load, not a video on one side and a game on the other, or temperature changes cannot be attributed to any one load. Second, watch long enough: a cycle may take several minutes to appear, and a one-minute glance can easily miss it. Third, do not change settings mid-log: a log made after a change belongs in a separate record, or before-and-after comparisons become meaningless. This record is useful twice over: it guides the user's own diagnosis, and it lets repair staff see the threshold gap and the cycle rhythm directly instead of relying on a verbal description such as "sometimes it spins and sometimes it stops".
Step Two: The Wider-Gap Test
Once the log is complete, the first comparison test is to widen the gap between the start and stop points. In concept, the change is simple: make the fan keep spinning until temperature has fallen further before it stops, or make it run at a low speed continuously at low temperatures instead of stopping at all. Option names vary by vendor software: some call it a fan-stop switch, some call it a custom curve, some call it hysteresis. Do not fixate on names; identify the function. Change one variable at a time: either widen the gap or disable the stop so the fan runs continuously at low speed, never both together, or it will be impossible to tell which change mattered.
If widening the gap clearly reduces or removes the cycling, the too-narrow-gap diagnosis is supported, and the result is itself a usable direction: keep a wider gap and accept the trade-off that the fan may run longer, or carry a little low-temperature background noise. If widening the gap changes nothing at all, take the other branch seriously: the temperature reading may be swinging too much, or the sensor driving the curve may differ from the temperature being watched. At that point, stop widening blindly and move to sensor-location diagnosis.
Step Three: The Single Fixed-Speed Control Test
This is the most decisive test in the guide. Set the fan to one fixed low speed, bypassing the curve and the thresholds entirely, and watch what the fan does under a fixed command. The two outcomes point in completely different directions. If the fan spins steadily at the fixed speed with no cycling, the fan itself, its power delivery, and its control path work normally at that speed, and the problem really is in the curve and threshold settings; the fix direction is then to keep a stable curve setting. If the fan still cycles irregularly or its speed jumps around on its own under a fixed command, the curve is not the main cause, and the problem may sit in the fan control or power delivery path; further curve changes would waste time, and the right move is to stop and seek inspection.
Observe a safety boundary while running this test. A fixed-speed test is for short diagnostic use only; do not leave the fan locked at a very low speed during gaming or heavy rendering, because inadequate cooling will let temperatures climb. Watch temperature throughout the test, and restore a normal setting at once if temperature rises clearly. After the test, whatever the result, restore a sensible curve; never leave a diagnostic fixed speed in place as a daily setting. A common mistake is to forget the restore step after a successful test, and then blame a later overheating episode on a new fault.
Sensor Location: Why the Temperature You See May Not Be the One the Fan Follows
A graphics card carries more than one temperature sensing point. Common examples are the graphics core edge temperature, the core hotspot temperature, and the memory temperature, and they measure different places and react at different speeds. Edge temperature moves relatively gently, hotspot temperature reacts faster to load changes, and memory temperature follows memory load. Which sensor the fan curve follows can change the result a great deal: a curve driven by a fast-reacting sensor will translate every small wobble straight into a fan command, making cycling more likely, while a curve driven by a gently moving sensor will stay steadier.
The way to tell is by comparison: show several sensor readings at once in monitoring software, and match them against fan on/off events to see which sensor's movement lines up best with the switching times. If one sensor has just crossed a threshold at the moment the fan starts, while the others have barely moved, that sensor is probably the driver. If several sensors rise and fall together, sensor location is probably not the main variable, and the diagnosis returns to the threshold gap. Note that not all software lets the user choose the driving sensor; where it cannot be chosen, this step still has value as explanation, because it explains why temperature can look low while the fan still cycles, and why editing an edge-temperature curve sometimes has no effect.
Boundary Cases: When the Threshold Method Does Not Apply
First, temperature rising steadily while fans never spin at all. That is not a too-narrow gap; it is cooling control not responding, and it should not be handled with slow curve experiments. Reduce load and seek inspection promptly. Second, irregular cycling that persists under the fixed-speed test. That result already rules out the curve, so further setting changes are pointless. Third, other case fans or a liquid cooling system running under a separate control scheme, where the whole case airflow interacts with the graphics card. Temperature wobble may then come from case airflow rather than the card itself, so fix the case fan settings first and re-log the card, or two moving variables will be impossible to separate. Fourth, a recent driver or control software update that changed the cycling rhythm. An update may have changed the default curve or the chosen sensor, so the first step should be to compare settings before and after the update, rather than assuming hardware trouble at once. Fifth, cycling accompanied by display glitches, crashes, or driver resets. Those symptoms go beyond fan control and should be recorded as a group for inspection, not handled as a fan-only issue.
Common Misconceptions
Misconception one: a fan that stops must be broken. Stopping at low load with low temperatures is the design; what matters is the cycling rhythm and the temperature position, not the single fact that it stopped. Misconception two: locking the fan at maximum speed counts as a fix. Maximum speed can hide the cycle, but at the cost of noise and wear, and hiding is not diagnosing; a genuine cooling problem later will then be harder to notice. Misconception three: changing every setting at once. Curve, fan-stop switch, case fans, and driver all changed together make it impossible to tell which change worked, so the next recurrence has to be diagnosed from scratch. Misconception four: the temperature shown by monitoring software is the only temperature. A graphics card has several sensing points, and judging from one alone can easily miss the sensor actually driving the curve. Misconception five: a spin-up sound proves bearing damage. A spin-up sound can simply be a stationary fan accelerating again, and steady operation under the fixed-speed test argues against a bearing explanation.
When to Stop and Seek Repair
Stop experimenting with settings if any of the following appears. Irregular cycling persists under the fixed-speed test; temperature rises while fans never spin; cycling comes with display glitches, crashes, or driver resets; or the fan shows clear mechanical trouble in any setting, such as a scraping sound or intermittent sticking. Bring three items to inspection: the temperature and on/off event log, the result of the wider-gap test, and the result of the fixed-speed test. Those three let repair staff start from the control path directly, without repeating tests already done.
Readers nearby are welcome to bring their records to HK GPU Fix for advice, at Room 1209, Dali Plaza, Shek Mun, telephone 2186 8010, WhatsApp 6558 6806. Describing the cycle rhythm, the temperature position, and which tests have already been run is enough for a quick judgement on the next step.
Supporting Checks: They Support the Verdict, They Do Not Replace the Decisive Tests
The three steps above are the decisive tests. This section covers only supporting checks: they remove interference and make those three results trustworthy, but on their own they cannot settle the verdict. First, observe under a steady load for a while: if the load itself swings up and down, temperature follows it and so does the on/off rhythm, so do not blame the thresholds until load fluctuation has been ruled out. Second, observe in a genuinely idle state: idle means no foreground programme is pushing the load, not merely "no game is running". Third, use only one fan-control programme at a time: two programmes writing to the same curve will fight each other, so disable all but one before retesting; the detailed classification of control-programme conflicts belongs to a separate article, and only the one-programme-at-a-time rule is used here. Fourth, check in passing whether intake or exhaust airflow is blocked and whether dust build-up is obvious: restricted cooling widens temperature swings and can indirectly make cycling more frequent, but cleaning is not itself a cure for a threshold problem, and cycling that persists after cleaning sends you back to the three decisive tests. Finally, a driver update can change the default curve preset, so cycling that appears only after an update calls for a before-and-after settings comparison before hardware ageing is blamed; but an update does not redraw the curve every time, and with no settings change the update should not be blamed.
Frequently Asked Questions
Does repeated on/off cycling under light load always mean a fault?
Not necessarily. Light-load temperatures naturally hover near a threshold, and a gap that is too narrow is enough to cause cycling with no component damage. The deciding evidence is the temperature and on/off event log, plus whether widening the gap improves the cycle.
What is the difference between a full stop and repeated cycling?
A full stop means the fan stays off while temperatures remain low, which is the silent design. Repeated cycling means the fan keeps switching between spinning and stopped, each time at roughly the same temperature. Only the second case needs this guide's method.
What is hysteresis, and why does a wider gap help?
Hysteresis means the on point and the off point are two different temperatures, with a gap between them. A gap wide enough means small temperature wobbles never trigger a switch, so the fan stays stable. A gap too narrow means every small crossing triggers another switch.
Why does cycling continue after I changed the curve?
There are several possibilities. The edited curve may follow a different sensor from the one actually driving the fan; the temperature reading itself may swing too much; or the problem may not be in the curve at all, in which case the fixed-speed test will show it.
Is the fixed-speed test risky?
It is suitable only for short diagnostic use and must never be left at a low speed during heavy loads. Watch temperature throughout, restore a normal setting at once if temperature rises clearly, and always restore a sensible curve afterwards.
Which sensor most commonly drives the fan?
There is no universal answer; it depends on vendor design and software settings. The method is not guessing but recording several sensors at once and matching which one's movement lines up best with fan switching times.
Can case fans affect graphics card cycling?
Yes. Case airflow changes the intake temperature around the card, which indirectly affects its temperature wobble. Fix case fan settings first during diagnosis, or two moving variables cannot be separated.
When should I stop and seek inspection at once?
Stop if cycling persists at a fixed speed, if temperature rises while fans never spin, if cycling comes with display glitches or crashes, or if the fan shows scraping sounds or sticking. Bring the test records along for inspection.