This is a ~2009 machine I purchased (cheaply) to use as a home server.

It is supposed to be fanless, but the power supply has its own fan, and there's an additional smaller fan on the side (far away from the CPU) apparently for hard disk cooling.

Everything is working as it should, but the thing is getting HOT.

Look here: https://www.guru3d.com/articles-page...-review,4.html
That's exactly how it looks in my case. The silver heatsink in the middle covers the CPU (I checked). It feels hot to the touch even when the machine is idle, with the case open.

I have configured 'sensors' with 'sensors-detect' and 'coretemp' reports on the CPU's temprature (there's another set of sensors for Nvidia's "stuff", those stay somewhere between 30 and 60 deg).

Running idle CPU temperature quickly rises to and stays at 85 deg Celsius.
Under stress it rises to ~90 deg Celsius - I stopped that after maybe 10min, didn't want to risk damage.

Whether the lid is open or closed doesn't seem to matter (I think when it's closed the hdd fan blows through the whole space, when it's open heat can dissipate upwards - both ways having roughly the same effect).

Anyhow, the spec sheet says the maximum temperature is 85.2 deg Celsius!
However:
I'd say that's the heat sink.
I'd also say that "at" the IHS is vaguely phrased.
Does 'coretemp' report the temperature "at" the heatsink, or somewhere closer to the CPU?

Any hints why this machine is getting so hot?
I also asked the previous owner, they haven't answered yet.

Heat is best imagined as a series of small resistors from wafer to plastic, through plastic, etc and your heatsink is at some unknown point in there. Here are some things you can do

1. Lift off the fan from heatsink, and dust or vacuum clean it. A paintbrush is good for getting between fins
2. Oil the fan & dust the blades.
3. Paint the heatsink matt black. It radiates heat better, worth 15º - 20º on it's own over a silver heatsink.
4. Read you sensors output carefully. Here's mine:

At 'high', the Bios or ACPI should have the fan flat out. At critical, it should be powering off in an orderly but rapid fashion. No hanging about waiting for things to behave. It's trying to save your silicon. The actual max junction at the wafer is ≅150º but even 120º on any of the silicon is basically disastrous. There's an effect called thermal runaway. The plastic around the wafer is ambient as far as the wafer is concerned. So when that gets to 120º (on a chip only specified 0-70ºC), it's way outside spec, the wafer can't cool and it just gets hotter until you turn the power off or it blows.

Nope, the IHS is the metal cover/"lid" of the CPU package itself. The heat sink sits on top of the IHS.
coretemp reads the temperature from a sensor inside the CPU itself, and will report a slightly higher temperature than if you measure the IHS, which in turn is slightly hotter than the heat sink.
Could be it's just designed that way. I have a Zotac mini-PC, and the fanless CPU gets way hotter than I'm comfortable with.

I'm no fan of passively cooled components that get much hotter than about 60°C. Why not put a small fan on the heatsink, perhaps one fitted with a resistor (or even better, thermistor) that prevents if from spinning at full speed? A fan would drastically improve cooling, and with speed reduction it wouldn't make much noise and would probably last just about forever.

I see.
That's precious information, and relieves me somewhat.

Does it also have an Intel Atom CPU?
_____________________

I have since re-applied thermal paste to both chips (btw I was wrong, the bigger chip is the Nvidia chip, I assume the smaller chip is the CPU - but they both feel very hot to the touch). It helps a little, a few degrees Celsius I'd say.
'coretemp' reports that the CPU starts off at 60 deg after boot and settles at about 80 after an hour or so.

I will have to experiment with fans for a while.

Another interesting thing is that the CPU does not throttle at all, it stays at maximum frequency at all times.

======================
Degrees celsius? Are you sure? Inside a completely dark case?
I'm sceptical.

In thermal theory,everything is compared with a 'black body' which is the ultimate in thermal dissipation. You will be skeptical until you do it.

I was on a 50HP 3 phase inverter project, throwing anything up to 400 Amps amps about at times, and the silver heatsinks which ran ≅85ºC under load; the physicist who took over the project had them painted matt black. I was as skeptical as you, but he did the research thing, measuring before & after.The temperature drop was 21-22ºC.

Even Motorbike shops here in the 1960s & 1970s (even into the 1980s) kept tins of 'cylinder head black' to cut down siezures on over-excitable 2 strokes of the period. It worked.

I have an ASRock board that is very similar to that board (looks almost the same except for colors, and both are Atom 330 CPU and nVidia ION equipped), and in the original condition it had a 40mm fan included for the CPU part (as in, from-factory) - I eventually had to replace that fan when it failed, but missing the fan does lead to the little thing running overhot when under load. I'd say add a fan - zipties or some other method can be easy enough. There are plenty of cheap 40-60mm fans available. It looks like it provides some headers by the ATX power area.

Re: the temperatures in general and damage, newer Intel CPUs (which I understand to be Pentium 4-era and newer) will thermal throttle rather than damage themselves when running overhot, but prolonged operation at maximum temperature can shorten the overall lifespan of the board (or at least surrounding components like capacitors). There's an old Tom's Hardware video floating around that shows this in practice with Pentium 4 chips with the heatsinks fully removed - they go right up to 85 or 90* C and 'keep going' at much lower performance, while older (e.g. Pentium 3) chips would usually suffer irrevocable damage. If memory serves not all of them 'throttle down' by reducing their clockspeed exclusively though - I vaguely remember something about the older chips just halting the execution pipeline for some % of the time to achieve a similar effect. Point being: it shouldn't be in any danger short-term, but it isn't a situation I'd leave be long-term.

Might be possible to start with top and htop to see load. Other metrics can show too.

There are some cpu tools to reduce cpu time/temps.

Might be a bios setting to reduce clock speeds.

At one time some of those sensors were not reading correctly. Can't remember what it was.

I'd think it must be running hot if the average use ought to be 1.2W. Guess I'd also check it directly on boot and see how long it takes to warm up..

Don't think there was ever a HLT issue on those or even linux.

^ thanks, jefro. Basic stuff which I'm sure I covered in my first post, but can't hurt to say it twice.

It's SOLVED - here's what:

I assumed this "nettop" was sanely designed, but it clearly isn't:

• the fan is for the hard drive, but it runs cool enough (~30 degrees Celsius) without it.
• the CPU, which clearly produces more heat than the Nvidia northbridge, has the smaller heatsink
• It's also too close to the lid - the heat build-up is considerable - the place just over the CPU is hot to the touch after a few hours

I have seen pictures of the same motherboard with larger heatsinks and fans on top of the CPU (and as obobskivich confirms).
In my case, apparently corners were cut to the point of insanity.

What I did:

• new thermal paste - can't hurt in a ~10y-old machine, brought temperatures down a few degrees. Not much.
• put the whole thing upright - that alone brought temperatures down another few degrees.
• took the fan out of it's place in the far corner of the case (as far from the CPU as possible!) and put it right over the CPU heatsink

The last one did it, the CPU idles (1% usage or so) at 60 deg Celsius now and even a stress test does not bring it over 80 degrees.

This is still more than in the test I read (link in 1st post) and I still notice that the CPU does not throttle. At all. Always the maximum 1600MHz.
That is while the system is idle! Yes, the server is running, but no stress. CPU usage just 1% or so.
@obobskivich, do you have any experience with that?

FWIW: I have two Atom systems - one with Atom 330 and one with Atom 525 (they're very similar dual-cores) - and they both sit at their respective 'spec' clockspeeds continuously (from memory, 1.6 and 1.86GHz respectively). I don't think they are meant to cycle clocks up/down like newer CPUs will. I've never had any issues with them overheating, but like I said one of them has a fan directly over the CPU, and the other one has a larger heatsink (reminds me of the heatsinks that Pentium IIIs used to have) and a fan mounted on the case that blows 'over' that heatsink. I remember my Core 2 Quad (which is similar-aged but much 'higher' in the lineup) only having very minimal 'clocking' itself - idle was 2.0GHz and load was 2.8GHz. Very different than modern systems that can cycle through a dozen or more turbo bins.

1 members found this post helpful.

^ Thanks again.
So the lack of throttling is not a Linux compatibility thing.
Still wondering though.
I'm still not getting anywhere near the temperatures shown here.
Mine:

I looked into this, and while you're not wrong the answer isn't that simple.
What I found out:

• If you have other hot things around the heatsink in question, a dark colour will help.
• Paint is too thick and might even have an opposite effect, you want anodising instead.
• Something about aluminium. The heatsink in question is copper though.
• Other factors play a major role and the effect of a properly anodised black heatsink might be disappointingly small.

https://electronics.stackexchange.co...on-a-heat-sink
https://superuser.com/questions/6712...ssipating-heat
https://www.edaboard.com/threads/hea...-silver.32753/

Practically speaking, I don't have a suitable replacement right now, and simply painting it black is not an option.

From my motor-bike days in the 70s, I have some expertise in this.

• Yes thyere is something about aluminium. The thermal conductivity is lower.
• It's also lower if the aluminium is 'shiny. You want rough edges, because they increase the surface area, therefore the efficiency.
• Sure, I said 'paint them.'I didn't mean 'paint them 15 times.'

OK.

That article (Guru3D) notes a peak of around 66* C. Something to consider is ambient tempreature - how warm/cold is your room/house/area vs theirs?

Broadly, I wouldn't consider 50-60* C ideal for a CPU at low load, but I wouldn't consider it 'dangerous' either, if that makes sense. What happens if you load it up?

Thank you for your interest.

The phrasing is really unclear there - the screenshot does not seem to relate to the text.
Also it says "it peaks around 66 degrees C", which I understand to mean that it goes up to 66 d C, and stays there, under load.

Warm summer days in a nordic country. Inside, it's a few degrees warmer than in winter I guess.

(that's 5 minutes with mostly CPU stress, some MEM and I/O too)
After a while it settles aroundMuch better than without the fan.

The machine is now in an open cupboard (not ideal but OK airflow), with a fan on top of the CPU.

Yours is running cooler I guess?

The Electronics Industry takes Ambient as 25º. I'm at a similar latitude to Norway/Sweden but we get a kinder weather as an island, , and 25º is a warm day for us. Heat is also measured in degrees above ambient, which makes more sense sometimes. So I'll bet you that 90º of the mil spec (=-55 to +125º) degrees actually survives working at +125º. The Industrial stuff (-40 to +85º) just about does.

1 members found this post helpful.