Quieter gaming PC: fan curves, airflow, dust

A gaming PC that sounds like a hairdryer is rarely a verdict on your hardware. More often it is a fan curve that panics at the first warm degree, helped along by a coat of dust.

The short answer

How do you make a gaming PC quieter? Work in this order: find out which part makes the noise, clean out the dust, switch the fans to PWM control and give them a flatter curve in the BIOS, and only then fix the airflow path. Buy fans or a cooler last. Each step has a manufacturer document behind it: Intel specifies how a four-wire PWM fan must behave [1], Noctua explains how to build a curve [2], and the Linux kernel documents the same controls for Linux [16][17]. We measured nothing ourselves and give no sound figures of our own. Every dB(A) number below comes from a manufacturer or a lab and says so. Our assessment: the curve is usually the quickest win, because default profiles tend to be steeper than they need to be [2].

How it works

1. Find the noise before you change anything. Dell lists the usual suspects: cooling fans, a hard drive that clicks or grinds, an optical drive, and a power supply that hums or buzzes [28]. Its method is simple: listen to where the sound comes from, then turn the computer off, disconnect non-essential parts, restart and see whether the noise persists [28]. Our assessment: a sound that swells and fades with load is almost always a fan, and a steady tone or a rhythmic click that ignores load points elsewhere. A clicking or grinding hard drive may be failing, so Dell advises a backup first [28]. Both retailer guides we read also name coil whine from the graphics card or power supply [33][34]. We found no manufacturer page that explains it, so we leave it at that mention. A short test helps: drop one fan group to its lowest setting for about ten seconds and listen [33]. Never leave a CPU, GPU or pump at zero for longer than that test.
2. PWM or DC: what the fan header actually does. A three-pin fan can only be slowed by lowering its supply voltage (DC control). A four-pin fan has a fourth wire that carries a pulse-width-modulated signal. The share of time that signal is "on" is the duty cycle, and 100 percent means full speed [2]. Intel's specification asks for a 25 kHz signal, with 21 to 28 kHz acceptable, and says that without any control signal the fan runs at maximum speed [1]. It also requires the fan's minimum speed to be 30 percent of maximum or lower, and describes three ways of handling duty cycles below that: Type A keeps the fan at its minimum, Type B does the same but switches the motor off at 0 percent, and Type C is a third variant for fans whose minimum equals their starting speed [1]. Noctua adds that most of its fans stop at 0 percent, some keep running, and it prints the value per model as "Rotational speed @ 0% PWM" [2]. Mixed setups work, with limits: a three-pin fan fits a four-pin header, and the board may switch to voltage control for it. A four-pin fan on a three-pin header runs at full speed unless the board supports voltage control [3]. Noctua also notes that an "Auto" setting sometimes fails to recognise a PWM fan and falls back to voltage control, so it recommends choosing "PWM" by hand for PWM fans [2]. ASUS says the same split in its own words: PWM mode for four-pin fans, DC mode for three-pin fans [13]. Gigabyte's recent boards use "hybrid" headers that detect which type is connected [15].
3. Calibrate, then draw the curve. Many boards offer a fan test that finds each fan's lowest working speed. ASUS calls it Q-Fan Tuning, says it takes two to five minutes, and warns you not to shut down or reset during it [14]. Noctua says some boards require this step before you can adjust speeds, and others do not [2]. A fan curve maps a temperature (usually the CPU's) to a fan speed. Noctua's view: default curves can be too aggressive, meaning too steep or too fast at low temperatures, and it prefers setting the curve in the BIOS or UEFI rather than in a program, to avoid software problems [2]. The ASUS BIOS manual offers the profiles Standard, Silent, Turbo, Full Speed and Manual, and also lets you set how fast the fan may step up and down, with options from 0 to 25 seconds [14]. Gigabyte lists a temperature interval setting, which it says can prevent fan noise caused by a sudden change of temperature [15]. This matters because Noctua points out that changes in fan speed can be more noticeable and distracting than a constant speed [2]. Our assessment: a CPU that jumps in short bursts makes a fast curve audible as constant revving. A slower step time or a flat plateau over the typical gaming range is calmer than a steep slope.
4. Know your guard rails. Intel defines Tjunction max as the temperature at which the processor starts to reduce power and frequency to protect itself, usually between 100 and 110 °C, and says you should look up the exact value on the product page for your CPU [26]. It also says that touching the maximum under load is not necessarily a concern [26], and that the processor throttles above it [27]. Noctua adds that running a CPU below its specified maximum can be considered safe [2]. Makers still differ on where the fans should hit 100 percent. ASUS suggests, for a manual CPU fan or pump profile, raising the duty to 100 percent once the CPU exceeds 75 °C, and warns that performance throttles otherwise [14]. Noctua's own example curve only reaches 100 percent at 90 °C and says a hotter CPU needs a more aggressive curve [2]. We do not give you one number. Take your CPU's own limit from its product page, keep the last curve point well below it, and test under a real load.
5. Linux: the same controls through sysfs. The kernel exposes fan control in /sys/class/hwmon/. fan1_input reports the speed in RPM, pwm1 sets the duty from 0 to 255 (255 is 100 percent), pwm1_mode selects DC or PWM, and pwm1_enable chooses the control method: 0 means no speed control (full speed), 1 means manual control through pwm1, and 2 or higher means automatic control by the chip [16][17]. Reading is allowed for everyone, writing only for privileged users [16]. Whether your board shows these files depends on a driver for its chip. CoolerControl's documentation states that undetected hardware cannot be controlled and that some drivers expose readings but no fan control [21][22]. For a script-based setup the lm-sensors project provides pwmconfig and fancontrol. pwmconfig tests every PWM output, briefly stops each fan for about five seconds, and the manual warns that this may raise processor temperature, so run it without load and check afterwards that all fans spin [20]. fancontrol then applies your settings: MINTEMP is the temperature below which the fan runs at minimum, MAXTEMP the temperature above which it runs at maximum, MINSTART and MINSTOP the lowest values at which the fan starts and still turns, and MINPWM is 0 if the fan may stop, or the MINSTOP value if it should never stop [18][19]. There is also an AVERAGE variable that smooths short temperature peaks [18]. If you would rather click, CoolerControl detects hwmon, liquidctl and GPU devices, offers graph profiles, and applies hysteresis and response-time limits through "functions" [21][23]. Its own wizard produces a starting point that you must verify under load [21].
6. Windows and the graphics card. We found no Microsoft page that describes a fan curve in Windows itself. Fan control sits in the BIOS, in the board maker's tools (for example MSI Center with "Cooling Wizard" and "Fan Tune" [35], or Gigabyte's Smart Fan software [15]) and in the graphics driver. The Windows power option "System cooling policy" has two values: Passive slows the processor before increasing fan speed, Active increases fan speed before slowing the processor [32]. It is a laptop-style trade-off and can be hidden on desktops. For AMD graphics cards, the Adrenalin software has "Fan Tuning" with a fan curve and a Zero RPM option that keeps the fans off under light load [24]. We found no NVIDIA page that explains fan control. The documentation of a third-party Windows fan program says that RTX cards limit outside control: a minimum of about 30 percent, no manual 0 RPM, and a zero-RPM mode that the card decides for itself [25]. Treat that as the program authors' statement, not NVIDIA's.
7. Airflow: a clear path beats a loud fan. Noctua's basic rule is that fresh air should enter at one side and leave at the opposite side, with intake fans in front or at the bottom and exhaust fans at the rear and top [5][6]. It recommends five fans as a baseline (three intake at the front, one at the rear, one at the top), accepts three as a minimum, and says more fans let each one run slower for the same airflow [5]. It also advises the largest fan size that fits, because larger blades move more air at lower speed [5]. The case matters as much: closed or glass fronts and tight dust filters restrict airflow, and Noctua suggests testing a filter by blowing through it with and without the filter [7]. Neat cables mean fewer obstructions and less turbulence [5]. Pressure is a trade-off. Positive pressure (more intake than exhaust) lets less dust in through gaps but can trap warm air, negative pressure expels warm air quickly but draws dust in through every gap, and neutral is the compromise [6][31]. Our assessment: for a gaming PC that sits on the floor or in a dusty room, slightly positive pressure with filtered intakes is the sensible default, and the right balance is something you confirm with your own temperature log.
8. Dust: clean the right way. Dell's instructions: turn the computer off, unplug it, and use compressed air [28][30]. Dell warns against vacuum cleaners and blowers because they create static that can damage components [29], and tells you to follow the instructions on the compressed-air can [29]. Noctua adds that you must hold the fan blades still while using compressed air, because a fan that spins up at high speed can be damaged [5], and its fan manual also advises against a vacuum cleaner [11]. As a rough guide Noctua suggests cleaning every few months, with environment as the deciding factor [5]. Dust on the filter and the radiator also makes the fans run faster for the same cooling, which is why a PC can get louder over time without any defect.
9. How to read a dB(A) number. Noctua's data sheet for one 120 mm fan lists 2000 rpm with 22.6 dB(A), and 1700 rpm with the supplied low-noise adaptor at 18.8 dB(A) [8][9]. At 20 percent PWM the same page gives 460 rpm [9]. These are the maker's figures, and Noctua's CTO notes that such figures are taken in free flow and that a fan can sound different once installed against a heat sink or radiator [10]. A lab that tested the same model with its own method listed a maximum of 31.3 dBA at 2134 rpm [12]. We do not know which differences in method explain the gap, and that is the point. Compare dB(A) values only inside one table from one maker, and read them as "this fan gets quieter when it spins slower", never as "this room will measure X". We cannot promise a result for your setup.
10. Undervolting and power limits: a pointer only. Lower voltage or a lower power limit makes less heat, and less heat means slower fans. AMD's tuning page lists the controls for its cards [24]. Intel warns that altering frequency or voltage may void warranties and reduce stability, security, performance and the life of the processor and other components, and tells you to check with the system and component makers [27]. This guide does not cover it. If you try it, change one thing at a time and check your maker's terms first.
11. When buying makes sense. Buy when the settings are exhausted and the problem remains. Examples: the fans still run fast at idle with a clean case and a sensible curve, a bearing rattles or scrapes, or the case itself blocks the airflow. Check the connector type (four-pin PWM), the size your slot supports, and compare data sheets from the same maker with rpm and dB(A) at the same conditions. We name no models and rank nothing.

A worked example: one curve, five points

This is a method, not a recommendation, and the numbers are a manufacturer's example. Noctua publishes a "gradual" case-fan curve with five points: 20 °C at 20 percent, 40 °C at 40 percent, 55 °C at 60 percent, 70 °C at 70 percent and 90 °C at 100 percent [2]. It says there is no universal curve and suggests adding about 10 percent to the values for CPU cooler fans [2]. Use it as a starting sketch:

• Read the idle speed first. In the BIOS, note each fan's rpm. On Linux, run ls /sys/class/hwmon/ and cat /sys/class/hwmon/hwmon*/name, then read the fan1_input file of the chip that belongs to your board. The hwmon numbers differ between machines.
• A tachometer signal gives two pulses per revolution, so Noctua's white paper converts frequency to speed as Hz × 60 ÷ 2. Its example: 86 Hz gives 2580 rpm [4].
• Set the control mode to PWM for four-pin fans, run the fan calibration if your board has one, and enter the five points.
• Replace the last point with your own limit. If your CPU's product page says Tjunction is 100 °C [26], a point at 100 percent well below that value is a more careful guard than the example's 90 °C, and ASUS suggests 75 °C for a manual CPU profile [14].
• Play a demanding game for about twenty minutes and watch CPU and GPU temperature and fan speed together. If the fans ramp up and down in short bursts, lengthen the step time or flatten that section of the curve [14][15]. If temperatures approach your limit, raise the curve there.
• On Linux you can test one channel by hand: echo 1 | sudo tee /sys/class/hwmon/hwmon2/pwm1_enable, then echo 100 | sudo tee /sys/class/hwmon/hwmon2/pwm1 (100 of 255 is about 39 percent) [17]. The chip path is an example. Afterwards set pwm1_enable back to the automatic value your chip documents, or reboot, and check that the fan responds.

Checklist

• Identify the noise source by listening and by turning one group of fans down for ten seconds, and back up your data first if a hard drive clicks.
• Update the BIOS and your system software only from the maker's own pages before you change fan settings.
• Set each fan header to PWM for four-pin fans and to DC for three-pin fans, or confirm that "Auto" picked the right one.
• Run your board's fan calibration if it has one and do not interrupt it.
• Draw a curve with a flat plateau over your typical gaming temperatures and a slower step time, rather than a steep line.
• Look up your CPU's maximum temperature on its product page and keep the 100 percent point well below it.
• Test with a real game for about twenty minutes and log temperatures and fan speed together.
• Clean the dust with the PC off and unplugged, use compressed air, hold the fan blades still and skip the vacuum cleaner.
• Check that intake and exhaust form one path from front or bottom to rear or top, and that filters and the front panel are not clogged.
• Compare data sheets from one maker only, with rpm and dB(A) at the same conditions, and buy only if the settings, the cleaning and the airflow did not solve it.

Related pages: CPU / processor; GPU upgrade: PSU, space and power connectors; PWM fan control.