Two machines at 40 dB(A) each are not 80 dB(A). They are 43.
Decibels do not add, because a decibel is a ratio on a logarithmic scale rather than a quantity. What adds is acoustic energy, so combining two sources means converting both back to energy, summing those, and converting the total back:
combined = 10 x log10( 10^(a/10) + 10^(b/10) )Doubling the energy is always 3 dB, whatever level you started from:
| Two identical sources | Combined |
|---|---|
| 30 dB(A) each | 33.0 dB(A) |
| 40 dB(A) each | 43.0 dB(A) |
| 45 dB(A) each | 48.0 dB(A) |
| 50 dB(A) each | 53.0 dB(A) |
And it keeps costing more machines to gain the same amount:
| Machines, 40 dB(A) each | Combined | Over one machine |
|---|---|---|
| 1 | 40.0 | +0.0 |
| 2 | 43.0 | +3.0 |
| 3 | 44.8 | +4.8 |
| 4 | 46.0 | +6.0 |
| 5 | 47.0 | +7.0 |
| 10 | 50.0 | +10.0 |
Ten machines are 10 dB above one. That is the entire scaling story, and it is why a rack full of quiet equipment can still be quiet while one loud thing in the corner ruins a room.
The loudest one sets the number
Here is the finding that should change what you buy. Put a 40 dB(A) machine next to a louder one and watch what the quiet one contributes:
| Quiet + loud | Combined | Added by the quiet one |
|---|---|---|
| 40 + 40 | 43.0 | +3.0 |
| 40 + 42 | 44.1 | +2.1 |
| 40 + 45 | 46.2 | +1.2 |
| 40 + 46 | 47.0 | +1.0 |
| 40 + 50 | 50.4 | +0.4 |
| 40 + 55 | 55.1 | +0.1 |
| 40 + 60 | 60.0 | +0.0 |
Once a source is about 10 dB below its neighbour it has stopped mattering, and at 15 dB below it has vanished. Silencing the quiet machine in a room with a loud one buys you nothing measurable.
This is the most common way money gets wasted on lab noise. Somebody spends an evening and a parts order making an already-quiet NAS quieter, in a room whose number is set entirely by the one machine with a small high-speed fan in it. The right move is always to find the loudest thing and deal with that specifically.
Distance beats almost everything
Sound spreads out, so the level falls by 6 dB every time the distance doubles. One source rated at 50 dB(A) at one metre:
| Listener distance | Level |
|---|---|
| 0.5 m | 56.0 dB(A) |
| 1 m | 50.0 dB(A) |
| 2 m | 44.0 dB(A) |
| 3 m | 40.5 dB(A) |
| 5 m | 36.0 dB(A) |
| 10 m | 30.0 dB(A) |
Now put those two facts next to each other, because together they answer the practical question.
Adding a second identical machine costs you 3 dB. Moving twice as far from the rack saves you 6 dB. Doubling your distance cancels three extra machines. Two identical 45 dB(A) machines heard from three metres come out at 38.5 dB(A), well below a single one of them at one metre.
Before buying fans, buy a longer cable. Distance, a different room, a different wall, or a cupboard, in roughly that order, will beat almost any component change you can make, and none of it voids a warranty.
The noise combiner does both parts of this at once: put in each machine’s rating and the distance it was rated at, say where you actually sit, and it reports the combined level and how much each source contributed.
What the numbers on a spec sheet mean, and do not
A rating in dB(A) is A-weighted, meaning the meter has been filtered to approximate how human hearing responds across frequency, discounting the very low and very high ends. That makes it the right single number for “how loud does this seem”, and it is why almost every quoted figure uses it.
Three things it will not tell you.
The distance it was measured at, unless the specification says, and without that the number means nothing. A figure at one metre and the same figure at three metres describe machines 10 dB apart.
What the noise sounds like. A-weighting collapses a spectrum into one number, so a smooth 40 dB(A) of airflow and a 40 dB(A) whine with a strong tone in it are the same figure and a completely different experience to live with. Tonal noise is the one people cannot stop hearing.
What your room does to it. Everything above is free-field arithmetic, which assumes sound spreads out and keeps going. A real room reflects, so the level falls more slowly with distance than the table says, and a small hard room can hold noise up substantially. Soft furnishings, a rug, a curtain and an open door all change the answer in your favour.
Treat the arithmetic as the floor of what you will hear rather than a prediction.
What a change actually feels like
Perceived loudness is not the same as level. The rough relationship most people work from is that 10 dB is about half as loud:
| Change | Energy | Perceived loudness |
|---|---|---|
| 3 dB | 2x | about 1.2x |
| 6 dB | 4x | about 1.5x |
| 10 dB | 10x | about 2x |
| 20 dB | 100x | about 4x |
Which puts the second machine in perspective. Doubling your equipment doubles the acoustic energy and makes the room about 20 percent louder to your ear. That is noticeable and it is not a disaster. Going from one machine to ten is twice as loud, which is.
It also sets a floor on what a modification is worth chasing. A change that buys 1 dB is inaudible in practice. If a fan swap is not worth at least 3 dB, it is worth doing for other reasons or not at all.
The part I cannot compute for you
Everything here is arithmetic on numbers you supply, and the numbers you supply are the hard part.
Manufacturer figures are measured under conditions that are not your desk, and a fan curve means a machine’s noise depends on what it is doing. An idle machine and the same machine under sustained load are different sources, and the load one is what you should be planning around, since that is when you are using it.
There is no measured number on this page and I have not put a meter next to any hardware. When I do, the benchmark database gains noise figures alongside the speed ones, taken at a stated distance, at idle and under load, because a single figure with no conditions attached is the thing this page has just spent several hundred words warning about.
FAQ
How loud are two servers together?
Three decibels louder than one of them, if they are identical and equally distant. Two 45 dB(A) machines make 48 dB(A). If one is louder than the other the answer moves toward the louder one, and once the gap reaches about 10 dB the quieter machine contributes almost nothing at all.
Why do decibels not just add up?
Because a decibel is a logarithmic ratio, not a quantity of anything. The quantity that adds is acoustic energy, and 3 dB is the label for twice the energy. Adding the numbers directly would say two whispers make a shout, which is why the combination has to go through the energy domain and back.
Is it better to buy quieter fans or move the machine?
Move the machine, nearly always. Doubling your distance from it is worth 6 dB, which is more than most fan changes achieve and costs a cable. A fan swap that buys 3 dB is a good one, and you would get the same result by stepping back about forty percent further.
My NAS is quiet but the room is loud. What do I fix?
Whatever is loudest, and probably not the NAS. If one machine is 10 dB above the others it is setting the number on its own, and improving anything else changes the total by a few tenths of a decibel. Work out which source dominates before spending anything: the noise combiner shows each source’s contribution separately for exactly this reason.
What is a reasonable noise target for a home lab?
That depends on the room and on whether anyone sleeps near it, so I would frame it by distance rather than by an absolute figure. Decide the level you want at the place you actually sit or sleep, then work backwards through the distance correction to what the equipment may be rated at. A machine that is too loud at one metre may be entirely fine at four.
Does a rack or a cupboard help?
Yes, and usually more than component changes, because a barrier and some distance work together. The trade is heat: an enclosure that blocks sound also blocks airflow, and a machine that gets hotter runs its fans faster and becomes louder. Ventilate deliberately, with a path that bends, since sound travels in straighter lines than air needs to.
Sources
- The energy summation and the 6 dB per doubling of distance are the definitions of the decibel and of spherical spreading. Both are computed on this page by the same code that runs the noise combiner.
- A-weighting, and the frequency response a dB(A) figure implies, is defined for sound level meters in IEC 61672.
- The relationship between level and perceived loudness, where roughly 10 dB corresponds to a doubling, follows the sone scale from Stevens’s work on loudness. It is a psychoacoustic approximation rather than a physical law, and individuals vary.
- Every dB figure on this page is an illustrative input, not a measurement of any product. Nothing here was measured with a meter. Take yours from the manufacturer’s specification, note the distance it was quoted at, and expect a real room to be less forgiving than the free-field arithmetic.