Long-time reader, first-time poster. I've just built a new gateway/webserver using the Jetway NC9C-550 motherboard (Atom N550 dual core), inside of a Mini-box M350 case. At the moment I have no active cooling.

The TDP of the processor is only 8.5W and the motherboard is fanless, and the M350 enclosure is designed for passive cooling.. I was hoping it would run just like that. but the first time I started up the box I watched the CPU temperature climb to 40C before I even got out of the BIOS menu so that didn't bode well.

I started up the box normally, installed FreeBSD no problem, and started copying 400GB of media over the network. It lasted about 1.5 hours at borderline temps before the system froze, I assume for thermal reasons. The enclosure definitely got warm to the touch.

Since the box is so small, it's going to be hard to put anything larger than a 40x40x10mm fan anywhere, but those are noisy.. Is there a genius passive cooling solution that I've missed googling? Any other ideas?

It could be that there's a contact problem between the CPU and the heatsink or that Jetway used a very inefficient chipset. But I doubt it.
And that case isn't the best fanless case according to some numbers posted here. But still: it should work fine without a fan.

Unless you're running it a fridge, there's nothing whatsoever to worry about 40C in the BIOS.
What you may not udnerstand is that these chips don't get much hotter at load. Atoms aren't like conventional CPU.
And you need to reach a certain difference in temperature with ambient for passive cooling to work. The gear should be designed for that. Intel says stuff like 80C for the VRM aera on the board is too much (the CPU itself can get hotter without issues of course).

The enclosure should get warm (how else is cooling going to work?). What's your definition of borderline temps? If it's like 50C, forget it: that's fine unless Jetway have done a terrible job and failed to warn you about it. That's not the kind of temperatures supposed to cause trouble at all.

In the bios all the idle states, that make the Atom so power efficient don't work.
Don't worry about 40 deg in the bios.

It's the other way around: Atoms are very efficient at load. At idle, they're nothing special in practice.

Thanks for the info, everyone.



I would call 65C borderline; I set the BIOS to do thermal throttling at 70 and shutdown at 75. I had not gotten proper temperature monitoring set up so I don't know what the system was running; I don't think the BIOS thermal throttling plays well with the OS so I assumed that was what crashed it.

It's good to hear that you guys think it should work. I will do some more testing.

What are the borderline temps you mention? I have to agree with the others, 40, 50 even 70 degrees should not be a cause of instability. A usual suspect could be the ram.

Are you able to run any benchmarks like prime or memetest? Don`t know if you can do that in freebsd, maybe you could on a linux live cd?

Here's an update:

First, to recap, I had the BIOS configured to do thermal throttling at 70C and to shutdown at 75C.

I ran memtest86+ while I went out to dinner last night; it completed one test without errors but the computer froze after about 2 1/2 hours. I restarted it and disabled the CPU throttling option in the BIOS. After that it ran 10 hours consecutively without freezing (completing 6 passes and detecting no errors in the RAM). I rebooted into BIOS where the reported system temperature was 69C.

Conclusions:
(1) thermal throttling in the BIOS does not work;
(2) operating temperature of the system is around 70C.

To me, that seems too high for an always-on system which requires high reliability ... do SPCR readers agree or disagree?

The atoms cooled passively idle high, i have Acer AH342 Windows Home server, has Atom D510 (lower clock that your D525), the cpu cooler is passive, the hole box only has one outake fan, it used to idle at 60C, i changed to faster Noiseblocker PWM and now im idleing at 50C or around there. Now in your case, you dont have any fans at all from what i see in the M350 case, so 70C seems reasonable for a case with no active fans, a hdd bracket that semi obstructs half of the vents on the top.

I think it's high. The CPU should be stable at 70C but what about the other components? I wouldn't worry about it too much if the tests show it's stable.
But I get temps around 60C with desktop dual-core Atoms which burn a lot more juice at load. It may be that the heatsink is inadequate, that there's too much and/or poor thermal compound or that Jetway botched something else. But this also confirms other reports according to which the M350 is not such a good case for fanless operation, at least if you use the bracket.

Maybe you could try a mod by replacing or adding heatsinks.
When building a custom case for a mobile Atom board, Morex used some kind of large thermal pad underneath the board (see SPCR's review). That and possibly a thin sheet of metal might help dissipate or at least spread heat.

An easier way to get lower temps might be to change the angle of the board to the direction of gravity. That can make a significant difference with fanless cooling depending on your heatsink.

@Abula: it's not a D525 which I would expect to run very hot without a large heatsink or a fan. It's a mobile CPU!

Yeah, I was thinking about trying to reorient the bracket so that the HDD sits over the IO end of the board..



The heatsink that comes installed is a pretty cheap all-aluminum model- you can see there are only 7 fins across. I bet something made with thin copper fins would do a lot better. Problem is it's a combination CPU+northbridge heatsink and I haven't found any immediate drop-in replacements. I could go for a separate northbridge cooler (like this?) and CPU cooler, but it also seems to be very hard to find heatsinks for atom CPUs. The motherboard doesn't even have a "socket" specification since the cpu is integrated.

Re: the Atom / Morex, you mean "energy efficient champ"? That's a nice looking system and a great writeup. Of course that's a single-core CPU.

I should do some power usage measurements. And I'll check the heatsink compound. I'd appreciate any suggestions for more effective CPU coolers.

I like this one: Nexus PSM-2500
or maybe this one: Dynatron I65G

Do you mean southbridge? I thought the northbridge was integrated in the CPU.


I think fins which are close to each other are suited to fanned cooling rather than fanless cooling. Look at Intel's fanless heatsinks. But you could probably improve them as well as the coupling with the CPU with better materials if nothing else.


If you search the forum, you'll find people who have done it with northbridge coolers and the like. Your CPU could well run a good bit cooler than the southbridge anyway.


Yeah, that's the one.
And it's an old CPU and an ancient chipset. Yours should by all rights run cooler but doesn't. Especially since your case looks like it has a lot more airflow than Morex's. My guess is that the thermal pad is a key part of the design or that Jetway has done something silly like pairing a single mobile Atom with a desktop chipset.


Yeah, and if you take out the heatsinks, look at what other chips could be heating it up besides the CPU.

My mistake. Yes, the southbridge.

Did some digging into this and according to Intel, the NM10 chipset is the only one compatible with the Atom N550. It's also got its own TDP of only 2.1 W.. pretty nice!

So last night I found the culprit.. a piss-poor thermal pasting job. I pulled off the heatsink last night and the thermal paste around the processor was like clay. Or foam rubber. Something non-conductive,anyway.

Scraped it off, cleaned the surfaces with isopropyl and added fresh heatsink compound. I ran the system with the cover off and the hard drive removed. Booted into memtest for 2 1/2 hours (2 passes). After that the BIOS reported the temperature a nice, cool 54C. That value didn't change much with the cover on.

Power measurements:

• 15.0-15.3 W in the BIOS;
• 15.3-15.5 W after adding the USB drive;
• 16.8 W nominal (17.8 W peak) during memtest.

These are measurements at the wall outlet so they include power lost in the brick and the picoPSU.

BTW the southbridge-to-heatsink interface is accomplished with some kind of foam. I'll post a pic tonight.

Your temperatures seem pretty reasonable for such a simple heatsink on a low-power machine. If you're willing (and you seem to be willing) to do some modding you should very easily get this computer cooled without the need for fans. As a comparison, I built a core 2 duo (well, Pentium Dual-Core E5200) system in the exact same case. It runs at about 20W idle @40C, 45W load @65C with the northbridge about 10C hotter. It's borderline, but it shows that you should be able to dissipate even three times as much power in that enclosure without the need for fans.

Here's a few photos of the system (article is in Dutch, sorry about that)

My turn to admit a mistake. This is wrong (I forgot they had integrated the graphics in the chip):

But there's something else than the CPU and southbridge which burns power. Intel's values for system TDP are not credible in light of the various measurements people have done. Desktop Atoms also use NM10 and burn about as much power as you do:

Unless your brick is unusually inefficient, there must be something (voltage regulation? I/O controllers outside the NM10?) burning power like a desktop on that board because Pineview latops reportedly use a lot less power. It would be interesting if you could post power consumption at idle and at load in an OS (try to stress the GPU too).


I got varying temps for identical Intel boards. I also suspect something like that in the case of the hotter ones.


That's strange. Are you sure?
If you're right the M350 is doing something right yet the temperatures people report with it are often pretty bad.

After continued observation, I have not seen the power consumption drop below a floor of about 15.3 W, even with aggressive CPU scaling. I think it must be the MoBo design, though there's not much there. Of course, that ultra-bright power button LED could be dropping at least a watt I have not had (and don't foresee having) an opportunity to push on the GPU, sorry.



Running under moderate CPU load for several hours inside the cover the temperature rose to a consistent 65 C; when I took the cover off it dropped pretty rapidly to around 54C, and down below 50C when CPU frequency scaling. A gentle fan over the heatsink gets the temperatures down into the 30s fast under moderate load. Still deciding what I'm going to do. But I agree the M350 is not the coolest case.

That system looks great! What heatsink is that?

It's an ASUS V70 without fan. Actually a pretty bad cooler, but it fits very well with the case and motherboard in my system. As with nearly all my computer stuff it's what I found second hand for a good price (I don't like buying new).

It's worth pointing out that AC bricks are not going to be better than 90% efficient and may be closer to 80%. (review here!) Plus, the efficiency seems to go down at lower demand. Finally the PicoPSU is rated between 85-94% efficient depending on the voltage and load level (datasheet). So 15.3-17.8 W at the wall * 0.85 (brick) * 0.9 (PicoPSU) means 11.7-13.6 W used by the system, maybe even less.

The processor at its rated TDP (8.5W), plus the southbridge at ITS rated TDP (2.1W) = 10.6W.

If we assume the MB uses a constant amount of power and that the full range of variation is due to the CPU itself: the high end of that (13.6 W) minus the power dissipation of the core (10.6 W) leaves 3 W burned by the motherboard, RAM, and HDD. There's not a whole lot of room to bring that down, though certainly in a laptop motherboard I would expect better.

Of course, the efficiency of the PSU varies with load, complicating the measurement process. Maybe it's time for SPCR to revisit its PSU analysis for AC adapters..

In your analysis the CPU consumes its TDP. There's *not a chance*. You'd have to load the CPU, the RAM and to do the craziest 3D. At idle, it should consume much less.
And obivously laptops consume a lot less.
The problem here is that we have an expensive mobile CPU on an expensive board with what is essentially desktop power consumption: viewtopic.php?t=57818
You don't even need a pico to get 18W idle with a cheap desktop board: http://www.silentpcreview.com/article1010-page4.html

I hope you get a lower power consumption idling in an OS. Otherwise that Jetway board is only marginally better than a desktop board which is a disapointment.

I would be tempted to install windows on it if only to take real time temp measurements. That would also get you more easily comparable power consumption figures.

As for your temperatures, do you see your system staying at this kind of load very often? A home server is likely to stay idle a lot of the time, a cpu that works at 70 degrees every now and then can still have a long lifetime. And considering your temps improved a little after replacing the thermal interface, you could be in safe territory.

You should get some further improvements by placing your case vertically, with some room underneath the bottom vents. Basically you want the fins parallel to the direction of airflow, which tends to be from bottom to top in a fanless system.

A Linux LiveCD works as well. It would be better to compare like with like but, assuming there's a substantial difference, I could compare Linux and Windows temps on the same NM10 board to give you a reference point if you need one.


It's worth trying as I said above but I wouldn'be so sure. Heatsinks with relatively tall fins can work in the horizontal position (depending on the case).

That's like hiring a butler to heat up a hotpocket!

You don't need windows for that. The measurements in my last post were fresh-off-the-sysctl-tree in FreeBSD. (in earlier posts I was rebooting into the BIOS, because memtest does not have a sysctl interface).


That is true- it would be more comparable to other systems on SPCR. But I'm only concerned with what mine will run. HFat has a good point that my TDP estimate is not realistic. I suspect the likely explanation for my power consumption readings is a low-efficiency AC adapter. 15W is still way better than my current tower (Celeron 3.20 GHz) which pulls down 80 W at idle!


I'm not too concerned about the temp, just disappointed. And a little concerned Playing around with the orientation is a good idea.

The bracket you are using to hold the hard drive can hold two 40mm fans. I believe you can install two of those brackets.

Noisblocker makes very good and very quiet high end fans. They have a 40mm x 10mm model.
http://www.noiseblocker.de/en/BlackSilent40.php
You can put one, two, three or four of them in and run them at 10 volts. They claim 11db at 12 volts. If you need more than two fans, you can
(a) move the hard drive to an external dock, or
(b) get one of those real small SSDs. It should probably fit into the hinged door on the front.

A more elegant solution in my opinion is to cut a 120mm hole on the top, mount a 120mm Nexus fan externally, and cover the fan with a Modright Filterright filter. Run the fan as slow as it will go, it will provide more than enough cooling.

What might be even more interesting would be to cut a hole for an 80mm Nexus or a 92mm Noctua and run that fan externally with a Modright Filterright filter.

I am running a passively cooled atom in an m350 case with an SSD as my server and have had no problem at all so far. Total silence is a little bit scary, you never know if the system has been powered on or not by simply listening to it.

Jetway has done a good job providing updates for the BIOS. I noticed one just the other day:
http://www.jetway.com.tw/jw/ipcboard_vi ... C9C-550-LF

"Solve the functions of shutdown temperature/throttling temperature work abnormally."

Not sure what that means in english, but it sounds like maybe they fixed some of the thermal issues?

I'm a couple versions back on BIOS simply because it is a rock solid server that never needs rebooting.

For what it is worth, my NC9C-550-LF eats about 26W (According to Kill-O-Watt), I've got a 7200k drive in there, a SSD for the OS(Debian Gnu Linux), a 120mm fan (really slow, but probably unnecessary), and an FX 300W 80Plus power supply (also with a fan).

sensors-detect finds Atom temp (coretemp) and Some Fintek chip which is supported in newer kernels.

Coretemp reports CPU at 47C

edit: 72->120mm fan, brain fart.