The Off-Key Bard
Technology

Why Does a Computer Fan Get Loud, and How Does It Know When to Speed Up?

Or: how a small plastic propeller becomes an automatic temperature-control system. What sounds like a machine groaning is a feedback loop at work.

Or: How a Small Plastic Propeller Becomes an Automatic Temperature-Control System

You're browsing the web and the computer is nearly silent, fans turning slowly enough to make a faint hum. Then you launch a game or start rendering a video, and within moments air is rushing through the case and the machine is noticeably louder. When the work ends, the fans eventually settle back down.

Nobody reached inside and flipped a switch. You never told the computer the processor was getting hot. It noticed, and adjusted. Behind that ordinary moment sits a small control system: sensors, firmware, a motor, and a surprising amount of aerodynamics. Moving air quietly through a crowded case is much harder than spinning a plastic propeller.

The Fan Doesn't Cool the Chip

Yesterday's post followed the heat out of the silicon: through the heat spreader, the paste, and into the metal fins of a heatsink. The fan handles the last step. It does not cool the processor directly, and it does not create cold air. Its job is to keep replacing the warm air hugging the fins with air that can absorb more heat. Without that, warm air pools around the heatsink and the heatsink can't release what it has collected. Even an enormous heatsink is only as good as the air moving through it.

The fan itself is a small brushless DC motor. Permanent magnets sit on the rotor, with the blades attached, and electromagnet windings sit in the stationary part. Control electronics energize the windings in a timed sequence that keeps the rotor turning, with no brushes making and breaking contact. That's compact, efficient, and durable, which matters for a part that may run for thousands of hours. It also means the speed can be commanded electronically, and that's the lever the computer pulls.

How the Computer Decides

Modern CPUs and GPUs carry temperature sensors across important regions of the die, and motherboards add more near voltage regulators and the chipset. That data goes to firmware. On many desktop boards, the BIOS or UEFI holds the fan settings: automatic tuning, preset profiles, minimum speeds, and custom fan curves. Once configured, it runs without any app open in Windows. Graphics cards typically carry their own sensors, firmware, and fan controller, so your GPU fans and CPU fan may follow different rules and respond at different moments.

A fan curve is the rule that turns temperature into speed. Picture temperature along the bottom and fan speed up the side. At low temperatures the line is flat and the fan idles. As the chip warms, the line climbs, and near a chosen limit it commands something close to full speed. A quiet profile keeps the fans slow until the processor is fairly warm. A performance profile ramps early. Neither is right in the abstract. The curve is simply a statement of which tradeoff the system should prefer, and as long as the chip stays inside its limits, both are reasonable.

Step back and the whole arrangement is a feedback loop. The processor makes heat, sensors measure it, firmware compares the measurement to the curve, a controller changes the motor's speed, the fan changes the airflow, the heatsink sheds heat faster, the temperature responds, and the sensors measure again. It's the same principle behind a home thermostat or the control system in a car, just running inside a PC case.

Why Fans Don't Just Chase Every Spike

There's a catch. A processor's workload can change almost instantly. Open an app, run a background task, or move between light and heavy game scenes, and a hotspot sensor can jump within moments. If the fan followed every wiggle, you'd hear it surge for three seconds, drop, and surge again all day. That's more irritating than a fan holding a slightly higher, steady speed, so controllers smooth their behavior.

One tool is hysteresis: using different thresholds for speeding up and slowing down. A fan might accelerate above 65°C but not decelerate until the temperature falls below 60°C. The gap stops it flipping back and forth when the temperature hovers near a boundary. Wikipedia's article on deadband describes the family of ideas. It explains that a deadband prevents "oscillation or repeated activation-deactivation cycles," and gives a furnace that turns on at 18°C and off at 22°C as an example of separate on and off points keeping a system from hunting.

Another tool is controlling how fast the response changes. Many systems ramp up quickly when heat rises sharply and ramp down slowly afterward, so the fan reacts to sustained load without overreacting to brief ones. The delays have to be chosen with care: too slow and temperatures climb needlessly, too aggressive and the fan becomes obvious.

Physics helps, too. A big heatsink has thermal mass. The die can spike within moments of a demanding calculation while the heatsink is still relatively cool, because heat takes time to spread through the base, heat pipes, and fins. The metal absorbs energy while the airflow catches up. Some liquid-cooled systems exploit this further by using coolant temperature, which changes gradually, instead of the chip's instantaneous reading, so the radiator fans stay calm through short bursts. Not every cooler exposes that sensor.

Voltage or Pulses

Most desktop fans use either three-pin or four-pin connectors. Both carry power and ground. A three-pin fan adds a tachometer wire that reports speed. A four-pin fan adds a dedicated control wire for pulse width modulation, or PWM.

A three-pin fan is generally controlled by varying the voltage it receives: less voltage, lower speed. That's simple, but brushless motors need enough power to start and stay running, so at very low voltages some fans stall or behave erratically, and different models have different minimums. That's why firmware often offers a calibration step to find a fan's usable range.

A PWM fan keeps a steady supply voltage and receives a separate control signal that tells its internal electronics how hard to drive the motor. The signal is described by its duty cycle: the fraction of each cycle it spends "on." The Intel four-wire PWM fan specification, in a copy hosted on mikrocontroller.net, sets the target PWM frequency at 25 kilohertz, with an acceptable operating range of 21 to 28 kilohertz. That is far above anything you could hear or see as motion. The name invites a bad mental picture of a fan that spins for a moment, stops, and repeats. It doesn't. The pulses are a command, not the motion; the fan's electronics and inertia smooth them into steady rotation. Duty cycle and speed also don't map exactly: 50 percent duty does not guarantee 50 percent of maximum RPM, since that depends on the fan's design.

The tachometer wire closes the loop. The fan emits electrical pulses as it spins, and the spec calls for two pulses per revolution, so the motherboard counts them to calculate RPM. That lets software display speeds and spot failures; a CPU fan commanded to spin that reports nothing may trigger a startup warning, and a stopped CPU fan can mean temperatures climb fast.

The tach only reports rotation, though. A fan can spin normally while its intake is blocked, the fins behind it are choked with dust, a cable is in the way, or a filter is too restrictive. RPM isn't airflow, and a spinning fan isn't proof of cooling.

Airflow Isn't the Same as Pressure

Fan specs usually list two numbers. Airflow, in cubic feet per minute, is how much air the fan moves. Static pressure, often in millimeters of water, is how hard it can push against resistance. They describe different situations. A fan facing an open grille hardly meets any resistance, so airflow is what counts. A fan pushing air through a dense radiator is working against a pressure drop, and it needs to keep moving air despite it. Fans sold as static-pressure designs use blade shapes, clearances, and frames aimed at that job, although the line between "airflow" and "pressure" fans isn't absolute.

Every fan has a performance curve. QATS's explanation of fan curves in electronics cooling describes it as the static pressure the fan can deliver as a function of flow, and says the operating point is where that curve intersects the system's own resistance curve. As the article quotes, "any given fan can only deliver one flow at one pressure in a particular system." So the maximum-airflow and maximum-pressure numbers on the box are endpoints of the curve, not values you get together. Your real performance lives somewhere between, set by your heatsink, filters, grilles, and cables. The same source notes a danger zone near the "knee" of the curve, where a small change in resistance, such as dust accumulating, can cut flow noticeably. This is also why clogged filters matter: the fans may spin normally and move less air.

Size adds tradeoffs. A 140 millimeter fan sweeps more area than a 120 millimeter one and can often move useful air at a lower, quieter speed, though mounting space and heatsink geometry matter too.

Why Faster Fans Get So Much Louder

Noise rises with speed, and it rises faster than you might expect. Faster blades disturb the air more energetically, increasing turbulence and the pressure fluctuations that radiate as sound. A paper from the Air Movement and Control Association on fan noise gives the standard scaling for fans of a given design: sound power grows by about 50 times the base-10 logarithm of the speed ratio. In that formula, a speed ratio of 1.6 produces a change of about 10 decibels, and a ratio of 1.26 about 5. Take a fan from 1,000 to 1,500 RPM, a ratio of 1.5, and you add roughly 9 decibels. That formula comes from industrial fan work, so treat the numbers as indicative for a small PC fan, but the direction is reliable. The fan isn't just moving more air. It's creating stronger and more frequent disturbances.

The rushing sound is mostly broadband, turbulence from blades, grilles, and obstructions. On top of it sit tones. The same paper describes fan sound as a broadband spectrum with peaks at discrete frequencies, tied to the blade-passing frequency and its harmonics. The blade-passing frequency is simple to work out: revolutions per second times the number of blades. A seven-blade fan at 1,200 RPM turns 20 times a second, so blades pass a given point 140 times a second, a 140 hertz hum. Because blade count and shape differ, two fans at the same RPM can sound quite different. As speed climbs, those tones shift upward too.

Decibels make this trickier to judge. The scale is logarithmic, and a rule of thumb says a 10 decibel increase is perceived as roughly twice as loud, so a fan rated 30 dBA is far noisier than one rated 20, not merely 50 percent louder. And ratings depend heavily on how they were measured, so numbers from different manufacturers aren't directly comparable.

Level isn't everything either. A smooth rush and a high-pitched whine at the same overall level feel very different, and people find tones more irritating than steady broadband noise.

When the Sound Is Mechanical

Not all fan noise is airflow. Bearings wear, lubricant degrades, and dust unbalances the rotor, producing ticking, grinding, or a low mechanical hum that no fan curve will fix. Vibration also travels: a fan bolted rigidly to a thin panel can turn the panel into a speaker, the same principle from the haptics post. Rubber mounts and well-braced panels help.

Quiet Is a System Property

The quietest computers aren't the ones with the fewest or slowest fans. A graphics card's zero-RPM mode shows one approach: when the load is low, its heatsink and natural convection can handle the small amount of heat, so the fans stop entirely, and start once temperatures cross a threshold. Even this needs hysteresis, or the fans would start and stop whenever the temperature hovers at the trigger point. Idle isn't zero heat. It's heat small enough to shed passively.

Case fans can be trickier to coordinate. If they respond only to CPU temperature during a game that hammers the GPU, they may stay slow while the graphics card dumps hundreds of watts into the case, which is why advanced tools can use GPU temperature or several sensors together. More fans aren't automatically better either; a well-planned intake-to-exhaust path with fewer fans can beat a crowd of fans circulating air in one pocket.

Laptops make everything harder. Their cooling sits in a thin chamber with small exhaust openings, so fans must spin very fast to force air through dense fins, producing sharp high-frequency noise. Many use centrifugal blowers, which draw air in and push it out sideways to build pressure for restrictive paths. The laws of airflow haven't changed. The space available has.

And a perfectly flat temperature isn't the aim. Modern chips run across a range and manage their own frequency and voltage, so a curve that holds a CPU at 55°C under every load may be louder than necessary, while 75°C under load can be entirely reasonable for hardware built to run safely there. Lower isn't automatically better. Appropriate is better.

When a computer is unexpectedly noisy, look at several numbers together: temperatures, fan RPM, utilization, and power. The cause is often a background task like updates or a malware scan that you didn't notice but the hardware did. One number rarely diagnoses anything.

The Bard's Take

A computer fan doesn't know what game you're playing, how many tabs are open, or whether you're rendering a video. Its controller knows only what the sensors report. Electrical activity rises, power and heat follow, a sensor notices, and the fan curve answers. The motor spins faster, the blades move more air, the heatsink sheds heat more quickly, and the temperature settles.

The details decide how it feels. PWM lets the computer regulate speed with electronic pulses, and the tachometer reports back. Hysteresis and response delays keep it from lurching. Blade geometry, fan diameter, pressure, and airflow resistance determine how well it moves air and what it sounds like, and that is why two fans at the same RPM can sound entirely different. A fan can spin perfectly and still fail to cool if its path is blocked.

So the best cooling system isn't the one with the most fans, the highest RPM, or the lowest temperature. It's the one that removes the heat the hardware produces, reliably, at a noise level you can live with.

Your processor performs billions of operations. Those operations produce heat. Sensors measure it. A controller decides how much air is needed. And a tiny brushless motor adjusts its speed thousands of times in a single session. You hear a fan getting louder. What you're really hearing is an automatic control system responding to physics.

Sources

← Back to the Desk — Technology