Running a Mac mini M4 Pro as a home server — the noise, and what I found trying to measure it

It stays on around the clock running agents and a local LLM. The fan barely moves off 1,000 RPM, and the temperature and power sensors report values that cannot be true.

The always-on computer in my house is a Mac mini M4 Pro with 48GB. Not as a desktop — as a server. Noise was the thing I most wanted to know about before buying, and once it arrived I measured it.

None of what follows is on the spec sheet. It is what you learn from leaving one switched on.

What it does all day

Three things on one machine.

  • Running development agents — reachable from outside the house, which is the whole point of it being a server
  • A Plex media server
  • Monitoring — the status of virtual servers running on a Windows box gets watched from here

On top of that, local LLM testing comes and goes. That turns out to be the only moment noise becomes an issue at all, which I’ll come back to.

Noise — the fan mostly doesn’t move

This is the question I’m asked most, so it goes first. The Mac mini has one fan, and I logged its speed continuously.

Fan speed — last 24 hours 998–1,003 RPM

Effectively fixed across the whole day. In this range I cannot hear the fan from anywhere in the room.

DeviceMac mini (M4 Pro, 2024) · 48GB · macOS 26.6 MethodNetdata, SMC fan sensor (F0Ac) at 1-second intervals
Fan speed — 7-day peak 3,581 RPM

The single highest reading in a week. Around 3.5× the resting speed, and at that point it is clearly audible.

DeviceMac mini (M4 Pro, 2024) · 48GB · macOS 26.6 MethodNetdata, SMC fan sensor (F0Ac) at 1-second intervals

Those two numbers explain the lived experience exactly. Most of the time the fan sits at around 1,000 RPM as though it isn’t running at all, and the only thing that lifts it is a local LLM. Those stretches are not long. Across seven days of logging, the peak was reached once.

Compared with the Windows machine that previously sat in the same spot, there is no contest on either heat or noise. That part is my own impression, though; for reasons below, I could not put a number behind the temperature.

The load is lighter than you would think

Three jobs running at once sounds busy. In practice the machine is idle most of the time.

CPU utilisation — 7-day average 5.7 %

2.9% user plus 2.8% system, with agents, Plex and monitoring all resident.

DeviceMac mini (M4 Pro, 2024) · 12 cores · 48GB MethodNetdata system.cpu, 7-day time average

Put another way: the M4 Pro’s CPU is overkill for this job. Which is also why the fan stays down.

And yet 48GB is not enough

Here is the part the spec sheet could not have told me.

The CPU is spare capacity; the memory is not. The moment a local LLM enters the picture the story changes. Models are held in memory while in use, so at 48GB the size of model you can load is what runs out first. With agents and Plex already occupying space, more so.

If I were buying again I would go for a higher-specified Mac mini. And “higher” means memory, not CPU cores. For this job, the same money is better spent on memory.

What I found trying to measure it — don’t trust the sensors

I meant to include temperature and power as well, and could not. The values were physically impossible.

The same monitoring tool reported this from the machine:

Metric Reported Why it can’t stand
CPU temperature (min) -1.5 °C A CPU cannot be below freezing
CPU temperature (max) 117.7 °C With the first one wrong, this has no basis either
Power (average) 254 W Far past what the Mac mini’s power adapter is rated for
Power (peak) 648 W Same reason

The cause appears to be the collector failing to filter macOS SMC sensor keys. Apple Silicon exposes a great many raw keys related to temperature and power, and some of them are not physical quantities a person can read. When a tool scoops those up and averages them, you get readings like a negative temperature.

The reason the fan sensor is fine is simple. There is only one of them, and the unit is RPM, so there is nothing to mix it with.

So this article carries fan speed and CPU utilisation only. Measuring temperature and power needs a different approach: a smart plug at the wall is the reliable way to get power, and temperature needs a separate macOS tool. I will write that up once I have actually measured it.

This is not a fault of the Mac mini. It is a problem with the combination of this monitoring tool and Apple Silicon. The same tool reports sane values from the Windows and Linux nodes.

In short

Question Answer
Quiet enough to leave on all day Yes. The fan sits at 998–1,003 RPM all day
When does it get loud Running a local LLM. Once in seven days, at 3,581 RPM
Is the CPU enough for a server More than. 5.7% average over 7 days
Is 48GB enough No. Memory runs out first once an LLM is on it
Buying again Same Mac mini, more memory
What about temperature and power I don’t know. The sensor values were not trustworthy, so they aren’t here

The worry before buying was noise; what actually became the constraint was memory. If you are looking for a computer to leave running in a living room, noise is not the thing to worry about. Decide the memory first, based on what you intend to put on it.

If you are setting one up for this, the settings worth checking on an always-on Mac are a separate matter, and worth doing before it goes on the shelf.