Undervolting to Cut Fan Speed

It is the only fix in this category that removes heat rather than moving it, and it is free and reversible. Everything else — better fans, a bigger cooler, dampening — deals with heat after it exists. Undervolting produces less of it.

A note on this guide: this rests on how modern parts are specified and how voltage relates to power, rather than on tuning done here. See how we test.

The problem, as it appears

The machine is loud under load and every available fix is a purchase. Better fans, a larger cooler, a case with more airflow, dampening material — all of them real, all of them addressing the same heat by shifting it somewhere else more efficiently.

Meanwhile the part generating that heat is very likely being fed more voltage than it needs.

Why more cooling disappoints as the only answer

Because cooling is a transport problem and heat is a production problem, and there is a limit to how well you can solve the first without touching the second.

A better cooler moves the same energy with less noise, which is worthwhile and has a ceiling. Past that ceiling you are buying diminishing improvements to the same transfer.

Reducing the heat at source moves the ceiling itself. Fewer watts to shed means lower fan speeds across the whole curve, which is a different kind of quiet from a fan that handles the same load more gracefully.

What actually matters

Default voltage is set for the worst sample, not yours

This is the part that makes the whole thing work, and it is not a trick.

Manufacturing produces parts that vary. To ship a model at a rated speed, the voltage has to be sufficient for the least capable sample that still passes, plus margin. An average part therefore receives more voltage than it requires, and a good one receives considerably more.

Power scales with voltage steeply, so a modest reduction produces a meaningful drop in watts — and watts are exactly what your fans are working against.

Undervolting is not underclocking

Worth stating plainly because the terms get used interchangeably and they are opposites in effect.

Underclocking asks the chip to do less work. Undervolting asks it to do the same work on less power. Clocks are unchanged.

On a part that was bumping into a thermal or power ceiling, undervolting can increase sustained performance, because it stops hitting the limit that was pulling clocks down. That is the outcome people find hardest to believe and it is a straightforward consequence of removing the constraint.

The failure mode is instability, not damage

The risk direction matters. Too little voltage means the chip cannot reliably complete operations, so it crashes or the driver resets. You lose unsaved work and you raise the voltage slightly.

You are not adding heat, not adding stress, and not operating outside a safe envelope — the uncomfortable direction is the other one. This is why it is a reasonable thing to suggest to someone who is not otherwise an enthusiast.

Stability is a week, not a benchmark

The common mistake is treating a passed stress test as the finish line.

Synthetic tests exercise a part in a specific pattern. A setting can survive an hour of one and still fail on a particular game, a particular encoder, or a workload that happens to hit a different unit on the chip. Real stability is a week of ordinary use.

Practical method: take a conservative reduction rather than the largest one that boots, use the machine normally, and if anything odd happens, give some back. The last few tens of millivolts are where all the instability lives and almost none of the benefit.

It stacks with the fan curve, and the order matters

Undervolting lowers the temperatures the curve is responding to, so a curve tuned before undervolting is now conservative for a machine that runs cooler.

Do this first, then set the curve — building a fan curve that is not the default covers the second half, and the step-down delay it describes is the other change that costs nothing.

The options, and who each suits

This page contains affiliate links. If you buy through them, Ugly Fixes may earn a commission at no extra cost to you. Products are selected independently.

Undervolting alone suits most people bothered by a loud machine under load, and it should be the first attempt precisely because it is free and reversible. If it works, nothing else on this page is needed.

Fresh thermal paste alongside it suits a machine several years old, where some of the heat problem is transfer rather than production. Paste degrades over years rather than months, so this is a supporting move rather than the main one.

Better case fans suit the machine still unpleasant once it is producing less heat. After undervolting, the same fans are running lower on the curve, so this may no longer be necessary.

A fan hub suits a machine with more fans than headers, which is the usual reason a curve cannot be applied properly in the first place.

The mistake to avoid

Chasing the largest reduction the machine will boot with. The stability that matters is not the stability of the first ten minutes, and the last increment of undervolt delivers a tiny thermal gain in exchange for an intermittent crash you will spend a fortnight blaming on something else.

The second mistake is undervolting to fix noise that was never thermal. A machine loud because of a restricted intake, a resonating panel or a failing bearing will be exactly as loud afterwards, and the diagnostic order in why a gaming PC gets loud is worth running first — undervolting is a good tool aimed at one specific cause.