Arctic replied to my support question and stated that they give the noise figure at maximum RPMs for all of their fans. I don't know if there are different standards for measuring noise levels for fans. If they all use the same techniques, that makes Arctic's number pretty impressive. I'm going to start with those since I was able to get a 5-pack direct from their web site (but fulfilled by Amazon) for US$30. I have also recently obtained a couple other 92mm PWM fans for comparison.

Noise ratings for things like case fans are given in "dBA", where the "A" indicates "acoustic profile". It's a weighted measurement intended to reflect how the human ear and brain perceive the sounds. (Back when Bell Labs was working out stuff for the phone system, they referred to this as "psychoacoustics".) The other common measurement, not interesting here, is "dBC", which is typically used for louder sorts of sounds where hearing damage is more the concern than perception. This article from Honeywell does a good job of explaining the difference.

I don't have anything like a proper acoustic measurement environment or equipment. Since I am interested in ballpark and relative differences, I did the cheap and easy thing.

• I have a piece of gear that can generate the 25kHz pulse train with PWM duty cycles configurable in 1% increments.
• I powered the fans directly from a +12v bench power supply.
• I bought the least expensive audio level meter I could find. (The seller sent me a lower-featured model, so it wasn't really a bargain.)
• I marked a position on the workplace and put each fan in exactly the same place when it was being tested.
• I set the audio level meter in "no timeout" mode, put it about a foot away from the fan position, and did not touch it at all for the duration of all the testing.

I was surprised to see that the background noise level in my basement "acoustics lab" was 44.4 dBA. I rechecked that several times during the testing, and it was pretty consistent. On the other hand, I could see the audio level meter jump around whenever anybody was walking around upstairs. And, if someone flushed a toilet, that put a temporary pause in the testing effort. The audio level meter was always jumping around at least a bit, so my measurement technique was to adjust the PWM duty cycle, wait a bit for the fan to change speed, and then "eyeball" what I felt was the average reading on the meter. Probably any difference of 0.2 dBA or so is not meaningful.

Other than the Arctic, I only had one of each fan. For the Arctic, they include cabling for chaining them together. I tested those with 1, 2, and 3 active Arctic fans since 3 fans is the target configuration for the project.

Here are the results, numerically (with 44.4 subtracted out) and graphically. You can find the exact models of the fans in an earlier project log.

I also measured each fan with no PWM control input at all (ie, that wire was disconnected). Each matched the 100% PWM duty cycle number, which is what the spec calls for.

Most of the fans did not show any noise difference until somewhere between the 30-40% PWM duty cycle. (See below for RPM measurements.) The exception is the Artic fans, which started changing between 20-30% PWM duty cycle, which matches the manufacturer's specs. Once the fans started reacting to the PWM changes, the correlation was roughly linear, except the Gelid, which rose rapidly and then plateaued. At the higher end, all of the fans were loud enough that they would be noticeable in a quiet bedroom. The non-annoying usable range is probably somewhere like 20-30% up to 60-70% PWM duty cycle. With 3 fans wired separately, there is the possibility of tuning them at different PWM duty cycles if that proves useful.

The noise winner was the be quiet! fan, with the Arctic a close second. The Coolermaster was not too far off, but it was similar to two Arctic fans. The Gelid was the loudest of all fans tested at all PWM duty cycles.

After I resolved a measuring problem, I ran another set of tests to measure fan speeds at various PWM duty cycles. For these tests, the PWM was generated by an ESP32, again in 10% increments. In an intervening experiment, I was able to see that those signals were very accurate. I measured the tachometer output of each fan with an oscilloscope that had a direct frequency read-out in Hz. The graph below scales the 2 pulses per cycle to RPMs.

It was pleasing to see that all of the fans were reasonably close to the spec values for minimum and maximum RPMs, and they had fairly good linearity in the expected part of the graph. The exception, again, was the Gelid, which reached maximum speed at 80% PWM duty cycle and was flat above that. The Arctic starts reacting at between 10% and 20%. The manufacturer says it's "absolute standstill 0 dB" below 40%. That clearly is not true, since the fan is turning slowly even at 0%, but it is pretty quiet down in that area.