GPU Fan Curve Setup Guide: Temperature–Speed Curves, Hysteresis & Balancing Noise Against Fan Lifespan
Key Takeaways
Your card's fans follow a temperature–speed curve in its firmware, which tools like MSI Afterburner can override. What matters most: keep the curve smooth with a little hysteresis, and remember it tracks the GPU's edge temperature — not the hotter hotspot reading. These choices shape both your card's noise and its fans' lifespan.
How GPU Fan Control Actually Works
PWM control and the VBIOS default curve
Modern graphics cards carry their own independent fan-control logic in the card's firmware (VBIOS), separate from the motherboard BIOS. A fan controller on the card follows a factory default curve built into that firmware, adjusting fan speed as the temperature changes.
Mainstream PC fans are speed-controlled in two ways. PWM (4-pin) fans keep the supply voltage at 12V and modulate speed using a PWM signal on the fourth pin; older 3-pin fans are controlled by varying the voltage (DC mode). When PWM is set to 0% on a fan that supports the design, the fan stops completely.
Zero RPM / fan stop: a designed feature, not a fault
Most modern graphics cards ship with a 0dB or Zero RPM mode: when temperatures are low — typically somewhere around 50–55°C, with the exact threshold differing by model — the fans stop entirely and the card runs silently. During idle, web browsing, or video playback, the heat output is so small that spinning fans are unnecessary.
Zero RPM brings side benefits: fewer running hours mean dust accumulates more slowly, the fan motors wear more slowly, and the card draws slightly less power (fan power draw is already small). One thing to note for troubleshooting: it is normal for stopped fans to spin back up automatically as temperatures rise. If the fans do not spin under load, however, the cause may be a mistaken custom curve, a driver problem, or a hardware fault — the sensible first step is to reset the fan curve to its default in software and test again.
VBIOS default curve vs. your software curve
When you enable "Enable user defined software automatic fan control" in MSI Afterburner, your custom curve replaces the factory default profile — but only while Afterburner is running in the background. Close it (without the right startup settings), and the card reverts to its factory curve.
There is a catch worth knowing. On NVIDIA cards, the software-controlled fan speed range is limited by the VBIOS minimum and maximum, so older Afterburner software curves could not truly reach 0 RPM (for example, the lowest they could go was around 30%), effectively disabling native Zero RPM. Newer Afterburner versions offer a hybrid option, "Override zero fan speed with hardware curve": the zero-speed portion of the curve is handed back to the native hardware curve, preserving the original stop behaviour. On Ampere-class cards there is a further subtlety — the GPU temperature, GPU power draw, and memory temperature act as three independent arbiters, any of which can wake the fans from a Zero RPM state.
Some modern GPUs (AMD Vega and newer, NVIDIA Ampere and newer) also support a "Use firmware control mode" that writes the curve directly into the GPU firmware, so Afterburner no longer needs to stay resident. It has limits, though: a cap on curve nodes (AMD cards allow up to 5 points across four slopes, NVIDIA up to 3 points across two slopes), and no step/linear mode switching or hysteresis adjustment.
Which temperature does the curve actually follow?
A modern GPU has several temperature sensors: Edge (average core temperature), Hotspot (the hottest single point on the die), and Memory Junction (memory temperature). When a fan curve uses "GPU temperature" as its horizontal axis, it generally means the core/edge temperature — the "GPU Temperature" monitoring item in Afterburner — not the hotspot.
The two readings can differ by more than 10°C, so confusing them leads to bad curves. On RDNA 2 cards the hotspot can sit 12–20°C above the edge reading, and one RTX 3090 sample showed an 11–14°C gap. Which sensor reports the hotspot can also vary with the workload. As a rough background reference, GDDR6 memory junction temperature has an upper limit of roughly 105–110°C.
Why the temperature matters: the NVIDIA temperature wall
On NVIDIA cards, boost clock is temperature-dependent. A temperature limit is commonly set at 83–84°C — reaching it causes the card to reduce clocks, with harder throttling and shutdown thresholds higher (roughly 89–95°C class, depending on model). One detail to get right: 83°C is not a "death line." Exceeding it only means the card downclocks; it is not instant damage. The real cost of chronic high temperatures is lost performance through throttling and accelerated ageing of components.
Hysteresis: Why a Small Delay Matters
Hysteresis is the gap, in degrees, between the temperature that starts the fans and the temperature that stops them. Without it, a temperature hovering 1–2°C around the threshold would make the fans switch on and off or constantly speed up and slow down.
A concrete example illustrates it: with a start temperature of 50°C and a hysteresis of 10°C, the fan only stops once the temperature falls back to roughly 40°C — it does not stop the moment the reading dips to 49°C. In Afterburner's Fan settings page there is a "Temperature hysteresis" option, and a commonly suggested value is 2–3°C, so the fans do not react to every tiny temperature ripple. Repeated rapid start-stop cycles are noisy, and they are harder on fan lifespan than steady running.
Shaping Your Curve in MSI Afterburner
Basic steps
The public, reproducible workflow is: open Settings (the gear icon), go to the Fan tab, tick "Enable user defined software automatic fan control," and the curve graph below becomes editable — horizontal axis GPU temperature, vertical axis fan speed percentage. Each white dot on the graph is one temperature-to-speed point; click a line segment to add a point, select a point and press Delete to remove it, and points interpolate linearly between each other.
Back up before you start. The first time you experiment, save the factory configuration into a profile slot (press the save icon plus a number key) so you can restore it at any moment. After editing, press Apply, then OK — the fans will then follow your curve as long as Afterburner is running.
Principles for drawing the curve
Keep the curve smooth and progressive, and avoid vertical jumps. An abrupt speed step (for example 35% at 65°C jumping to 80% at 66°C) makes the fans constantly accelerate and decelerate, and the human ear is more sensitive to that wavering than to a steady wind sound.
A typical curve structure is: keep quiet in the idle/low-temperature zone (0% or low speed) → ramp gradually through gaming temperatures → become more aggressive in the high-temperature zone near the temperature wall. Every specific temperature–speed pairing below is an example only — every card is different:
- Example pairing A: 25–30% at 35–40°C, 40% at 55°C, 55–65% at 70°C, 75–85% at 80°C
- Example pairing B: 40% at 55°C, 55% at 70°C, 70% at 80°C
These are starting points from published guides, not standards. Adjust them against your card, your case airflow, and your noise tolerance, and re-test with your own measurements.
Two principle-level boundaries are worth respecting: above 70°C, do not set the speed too low, especially in a small case or a system with poor ventilation; and from 78–80°C upward, use a more aggressive speed to avoid hitting the temperature wall and throttling.
Making the curve survive reboots and driver updates
The software curve is not permanently written to the hardware (unless you use firmware control mode). To keep it applied, you need both: Settings → tick "Start with Windows," and then in the main interface press "Apply at Windows Startup" (the Windows icon). Enabling only one of them can leave the factory curve back in place after boot.
Settings can be silently reset after driver updates, so after each update, re-open Afterburner, check the curve, press Apply, and re-save the profile. A common reason a curve does not take effect after reboot is that Afterburner launched before the GPU driver had finished initialising — the fix is a startup delay via the Task Scheduler, or using Afterburner's own Start with Windows option, which waits for the driver. Do not run multiple automatic fan-speed tools at the same time (MSI Center, vendor utilities, and the like), because they overwrite each other. And expect settings to be lost when you swap graphics cards — screenshot or write down your curve.
The Trade-offs: Noise, Wear, Dust, and Temperature
Why running 100% all the time is a bad idea
Speed and noise do not scale linearly: beyond a certain RPM, noise rises sharply while cooling gains flatten out — diminishing returns. Sustained high speeds also increase bearing friction and heat, accelerating wear. As a rough public reference (ratings vary with bearing type), typical GPU fan lifespan figures cited in open discussions are around 10,000–30,000 hours for sleeve bearings and 30,000–100,000+ hours for ball or hybrid bearings. Finally, the faster and longer fans run, the more dust they pull in, and a dust-caked heatsink degrades cooling efficiency in turn.
The exception is short: a brief 100% run to probe a temperature ceiling is a legitimate test. Living at 100% day-to-day is wasteful — all noise and wear for little extra cooling.
Why 0% under load is worse than it sounds
Fans that do not spin under load invite overheating — temperature-wall throttling at best, protective shutdown at worst. A curve set too conservatively in pursuit of silence also lets the hotspot and memory temperatures run away: remember, the curve tracks the edge temperature while the hotspot can be 10–20°C hotter, so the hotspot gets into trouble first when speeds are inadequate.
One diagnostic rule of thumb: if 90–100% fan speed still cannot hold the temperature down, the problem is usually not the curve — it is case airflow, a dust-clogged heatsink, or aged thermal paste.
The human ear cares about smoothness
People are more sensitive to fans speeding up and slowing down than to a constant airflow sound — which is why a smooth curve plus hysteresis matters more than simply choosing "a low speed." In summer, when ambient temperatures are high, consider shifting the whole curve slightly more aggressive (lowering the start threshold or raising the speed at each point) to compensate.
Cooler Designs Change the Rules: Blower vs. Multi-Fan Axial
Not all cards should share the same curve shape. A blower (single-fan) card and a dual/triple axial-fan card are physically different beasts:
| Blower (single fan) | Axial (dual/triple fan, AIB mainstream) | |
|---|---|---|
| Airflow direction | Air is drawn in and exhausted out of the case through the rear I/O bracket | Air is blown down through the heatsink and exhausted into the case |
| Strengths | Hot air leaves the case directly; suits small cases and poor airflow; stacked cards do not blow hot air at each other | Large heatsink area and multiple large-diameter fans can run at lower RPM — cooler and quieter at the same class |
| Weaknesses | One small fan needs high RPM for enough pressure — generally louder and warmer | Hot air stays in the case, so it depends on case airflow to carry the heat away |
| Curve implication | The noise penalty in the high-RPM region is steep; treat the curve slope with care | More tuning headroom; a gentler curve can be used |
Practical consequence: do not copy an aggressive curve from a triple-fan card onto a blower card — 60% speed might be whisper-quiet on one and roaring on the other. That is a physical difference; no specific decibel figures are cited here because no source supports particular numbers.
Validating Your Curve: Real Games, Not Power Viruses
- Watch temperature and speed in real time. In Afterburner → Settings → Monitoring, tick GPU Temperature and Fan Speed, enable "Show in On-Screen Display," and play an actual game to confirm the curve behaves as drawn.
- Give temperatures time to settle. After each change, run the same game and the same scene for 15–20 minutes before comparing — temperatures need time to reach equilibrium, and a two-minute reading tells you little.
- Treat FurMark with suspicion. FurMark deliberately pushes the GPU to power and temperature extremes in ways real workloads never do — manufacturers have called this class of load a "heat virus" or "power virus." Modern cards have throttling protection, so it will not normally kill a card, but it does not reflect real gaming load; running it for long periods mostly stresses the power delivery and adds wear. Real games, or realistic loads like 3DMark or Unigine benchmarks, are the better way to validate a curve. If you use FurMark at all, keep it as a short temperature-ceiling check.
- If throttling or instability appears, suspect temperature or power delivery first — do not just keep adding fan speed.
A Note on Undervolting
Undervolting lowers power draw at the same clock speed, which reduces heat — so the same fan curve ends up running at lower temperatures with quieter fans. The two techniques complement each other. Our site already has a dedicated undervolting article, so this guide mentions the relationship only in passing rather than re-teaching it.
FAQ
Q1: My fans stop spinning when I am just browsing. Is the card broken?
Almost certainly not. Zero RPM is a designed feature: at low temperatures (around 50–60°C, depending on the model) the fans stop completely. The judgement criterion is temperature — idle and cool with stopped fans is normal; hot under load with stopped fans is the actual problem, and the first step is resetting the fan curve to default and re-testing.
Q2: Should I copy someone else's curve numbers from the internet?
No. Every card has a different cooler, every case has different airflow, and everyone's ambient temperature and noise tolerance differ. Published numbers are starting points only — you must measure and fine-tune on your own hardware.
Q3: Why does my GPU report two different temperatures, and which one does the curve use?
The card has multiple sensors: the edge (average core temperature) and the hotspot (the hottest single point), plus memory junction. The fan curve's temperature axis generally follows the edge temperature — the "GPU Temperature" in Afterburner — while the hotspot can run 10–20°C higher. Designing a curve around the edge reading while ignoring a runaway hotspot is a classic mistake.
Q4: I set my curve in Afterburner, but after reboot it is gone. What happened?
The software curve lives in Afterburner, not in the card (unless firmware control mode is used), so it must be re-applied at startup: enable both "Start with Windows" in Settings and "Apply at Windows Startup" in the main window. Also check after every graphics driver update, which can reset the behaviour — and do not run a second fan-control tool (MSI Center, vendor utilities) that will fight Afterburner for control.
Q5: Is FurMark a good stability test for my new curve?
It is a power-virus class load that does not represent real usage. Modern cards protect themselves, so it is unlikely to destroy hardware, but a long FurMark session mainly tortures the power delivery and adds wear without telling you how the curve behaves in real games. Use a real game session of 15–20 minutes, or a realistic benchmark, and reserve FurMark — if at all — for a short temperature-ceiling check.
Q6: My card hits high temperatures even at 90–100% fan speed. Should the curve go higher?
If the fans are already near maximum and temperatures will not come down, the curve is not the problem. Look at case airflow, dust in the heatsink, and aged thermal paste instead.
Q7: Does the "Override zero fan speed with hardware curve" option matter?
Yes, if you value silence at idle. A plain software curve on some NVIDIA cards cannot reach true 0 RPM because the VBIOS caps the software range, which silently kills native Zero RPM. The hybrid option hands the zero-speed portion back to the native hardware curve, preserving fan-stop while the rest of your custom curve stays in force.
When to Seek Professional Help
A well-tuned curve solves most noise-and-temperature complaints, but some symptoms point to hardware: a fan that rattles or grinds at any speed, a fan that never spins even under load after the curve is reset to default, or temperatures that keep climbing no matter what the curve does. Those call for hands-on diagnosis rather than more curve tweaking — fan replacement, a deep clean, or repasting may be needed.
If you would rather have a technician look at it, HK GPU Fix offers professional inspection and chip-level GPU repair. Our three-tier fees run HK$400–2,800, with the final quote confirmed after inspection. Reach us on WhatsApp at 6558 6806, call 2186 8010, or visit the shop at Room 1209, Dali Plaza, Shek Mun.