CPU cooler makers publish decibel figures for almost every model, which puts them ahead of most appliance brands. The problem is what the figures describe.
We read twenty four specification sheets from eleven brands, covering air towers, low profile coolers, a fanless heatsink and all in one liquid coolers. Nearly every figure is a maximum: the cooler at 100% PWM, with the fans flat out. Your CPU spends most of its life nowhere near that.
Our 7 picks are listed above, with full specifications for each.
Look at the air cooler maxima in this guide and they land between roughly 22 and 29 dB(A). Noctua's NH-U12A publishes 22.6, Thermalright's Peerless Assassin publishes 25.6, DeepCool's AK620 publishes up to 28, ID-COOLING's Frozn A620 publishes 27.2. These are different coolers with different fans and different fin stacks, and on paper they are almost the same machine.
That is not because they sound alike in your case. It is because a maximum speed figure measures the least interesting state a cooler ever reaches.
Two makers publish something better
be quiet! publishes three figures for each cooler, at 50, 75 and 100% PWM. For the Dark Rock Pro 5 those are "8.9 / 16.3 / 23.3". The gap between the first and the last is 14.4 dB, from one cooler, depending only on load. Comparing a 50% figure to another brand's maximum tells you nothing, but comparing be quiet! coolers to each other works: the Dark Rock Elite publishes 11.0 / 19.5 / 25.8, so the cheaper Pro 5 is quieter at every point.
DeepCool does something different and just as useful. The ASSASSIN IV has a physical switch, and DeepCool publishes a cooler level figure for each position: up to 22.6 dB(A) in Quiet Mode and up to 29.3 dB(A) in Performance Mode. You can see what the switch costs before you buy it.
be quiet! has a speed switch too, with Quiet and Performance positions listed in its own specification table, but it publishes only one set of noise figures and never says which position they describe. Two makers, the same hardware idea, and only one of them closes the loop.
Nobody says where the microphone was
Not one of the twenty four sources named a measuring distance. Not one named a standard such as ISO 7779, and none described the test room.
This is the same problem we find with appliances, and it means a figure from one brand and a figure from another are not really comparable, even when they look like they are. Within one brand's range they are useful, because the test was presumably the same. Across brands, treat a 2 dB difference as noise in the data rather than a difference you would hear. For the longer version, see why you cannot compare dB ratings across brands.
Where the specification sheets go wrong
Three things we found that should make you read the number, not just the badge.
A copied block. Thermalright's Peerless Assassin 120 SE and Phantom Spirit 120 SE use different fans at different rated speeds, 1550 and 1500 RPM. Both pages publish "25.6 dBA", "66.17 CFM" and "1.53mm H2O". One of those blocks has been copied from the other.
The same pump, rated twice. Cooler Master publishes "Pump Noise Level 12 dBA (Max)" for the MasterLiquid 360 ATMOS White and 25 dBA for the 240 Atmos, with the same pump dimensions, the same MTTF and the same power draw.
No figure at all. ARCTIC publishes no acoustic figure for the Freezer 36 or the Liquid Freezer III 360, on the product page or in the downloadable spec sheet, and only the older Freezer 50 carries one, in sones. ENDORFY publishes a full table for the Fortis 5 with no noise row in it.
The cheapest quiet upgrade is in the box
If you take one thing from this guide, take the Low-Noise Adaptor. It is a small cable Noctua includes that caps the fan speed, and the figures move a long way: the NH-L12S goes from 1850 RPM and 23.9 dB(A) to 1400 RPM and 16.8 dB(A). The NH-U12A goes from 22.6 to 18.8.
The general rule holds beyond Noctua. Buy more cooler than you need and run it slower, and you will get a quieter machine than buying the cooler with the lowest number on the box and running it hard.
We did not measure any of these coolers. Every figure here is the maker's own, quoted as published, with the condition attached wherever the maker attached one.
