If it comes to Mean Well bricks , my company orders larger quantities from time to time , so let me know , we order a bit more and send them , whoever interested.

We usually order GS60 (12V version) , GS90 (12V) and GS90 (19V) .

Also let me know what tip You want . They come with 5.5/2.1 tip , but we change them to 5.5/2.5 tip. GS90 variant comes with 5.5/2.5 tip as a standard.
It is possible to get 7.4/5.0 tip as well (for Intel DH61AG board) .

Actually the main reason I did the Win7 power measurements was to be able to directly compare them to Linux power measurements on my setup, so to avoid comparing my Linux measurements with other peoples Win7 measurements.
Ultimately what I'm interested in is getting the same power draw with Linux as with Win7 (in the same circumstances).


I actually think (but if any electronics expert is reading this please correct me if I'm wrong) that incandescent light bulbs with a known rating make very good known loads due to their stable power draw and the PF of 1. I don't have a 12V light bulb here right now but I should be able to get one in the next few days.


I had no intention to imply that you wanted to fool anybody, I only asked you for more details so that we could try to find the cause of the difference between your and my readings.

Thanks for the details, the fact that you had the Ethernet cable unplugged explains a lot since that reduces power draw by approx. 1.5W in my testings.

You say you use wireless, but I guess you did not have a wireless adapter connected to the DN2800MT during your power measurements?

Also did you have "OS ACPI C3 Report" and "PCIe ASPM Support" enabled in the BIOS?

Was you monitor in DPMS standby (triggered by Win7 after 10mins of idle) or was it still showing a picture when you had the 7W reading?

I'm interested in this board too, but i've read many complains about video output issues.
The Win7 graphics driver was updated on 2/29/2012, could you describe better which issues you are getting with HDMI ouput?

Obviously. But I was looking at your earlier MeeGo measurements and I'm not sure how to relate them to your later measurements in terms of BIOS settings and peripherals.
For clarity, could you state what difference you get between Win7 (with official drivers), CentOS6 (with generic drivers) and MeeGo (with official drivers). Is it about the same betwen Win7 and MeeGo and an extra 2W for CentOS6?


I don't know how much variance there is between bulbs and how long you need to wait for them to heat up but the PF is what I thought might make them unlike a computer (better to determine the ideal efficency but I assume worse for our purpose).

You can find them on http://se.rs-online.com but the price is not cheap at all. Here in Italy I found an official reseller for meanwell who is selling the GS60A12 for about 23€ (without VAT I suppose) when RS sell the same for about 44€ ...

They are not comparable, I was using different and varying hardware setups when I was doing the first Meego, Centos, Fedora tests (no SSD, instead CF card with SATA>Cf adapter, 2/4GB RAM, live boot from USB stick, etc...).

I still have to redo tests with Mageia, Centos, Meego, with exactly the same hardware configuration as I used with Win7, but I haven't had time for that yet.


You might well be right, but in that case I have no idea what I could use as a good known load.

I got the board in the mail and set it up. I've installed Win7 32-bit and all of the Intel drivers. Everything looks good, unfortunately I'm getting about 15W idle. I really think it's the 12V power brick I'm using, which I bought for my 120W PicoPSU and found it to be one of the few bricks that were 120W+ that didn't require a fan. It's probably not the most efficient thing.

Does anyone have a suggestion for a relatively efficient power brick that is say $25 or less? Thanks!

Looks like people recommend the Mean Well. Does the GS60A12-P1J have the right connector to work with the DN2800MT?

From DN2800MT product guide:


From GS60 Specification pdf:


I guess therefore that they are not compatible in the basic configuration of the GS60 but probably you need to ask the seller for the characteristics of the model that is selling.
Anyway I think you can find adapters in electronics store.

No wifi during testing , OS ACPI and PCIe ASPM enabled. Monitor was showing picture . Nothing changed when in standby.

Mean Well GS60 comes with 5.5/2.1 as standard hence gs90 has "proper" 5.5/2.5 tip . Soldering different tip is difficult because cable itself is thick. We use third party company to solder 5.5/2.5 tip.

Thanks. Got the GS60 for $20 and plug adapter for $1 on eBay. Looking forward to seeing if this improves my power efficiency

I have now repeated all tests I did with Win7 using exactly the same setup with Centos 6.2. Centos 6.2 doesn't have any specific driver or optimizations for the Cedarview Atom, so predictably it draws more power. The difference varies between 1.5 and 2.4 Watts depending on the specific test, which IMHO is a realistic difference due to the lack of GPU drivers.

I decided to make a whole article out of the tests, so you can find it with the results here:
http://www.linuxtech.net/reviews/intel_DN2800MT_cedarview_atom_power_draw.html

Please note that the article is still incomplete and I still have to do the tests with Mageia and Meego so it's not yet officially "published", but I thought I might already give you the link so you can see the Centos test results.



Thanks for the info. In that case your test is comparable to my 'Test 1c' setup (see article at link above) where I measured 8.2 Watts, I guess the difference of 1.2 Watts compared to your 7 Watts result can be easily explained by the different watt meters and PSUs.

thanks linuxman,your review is really a great job.

I have been looking into how the gma3600 drm kernel driver (actually called gma500_gfx) in the latest Linux 3.3rc7 kernel should be used, since it wasn't working out of the box with Mageia and while doing this I think I found a bug with either the gma3600 drm kernel driver or the BIOS of the DN2800MT.

Basically the kernel driver always detects the presence of a 1920x1080 LVDS panel, despite I have nothing attached to the LVDS connector. I have mailed the developer at Intel whose email address appears in the kernel module and hope that this will lead to a working solution.

Right now with this bug as far as I can see the Linux kernel driver module is not usable with the DN2800MT (it might work fine in Cedarview netbooks that do actually have a LVDS panel attached).

Finally got my Meanwell brick back (GS60) , and dropped 1W from IDLE State . Now I have got firm 6W (sometimes jumps to 7W , but most of the time stays at 6W) . Max CPU using LinX is at 9W now.
Previous measurements using generic brick and setup can be found in first post in the thread.

I can also confirm that newest graphics drivers from 8th march fixed problems with display (flickering screen)

To linuxman : I am pretty sure that 1.2W difference is a kind of measurement error . I need to get same powermeter as Yours , mine is some cheap meter and it may cause different readings.
I will say it once again , despite any numbers. This board shocked me in terms of power usage: 6,7 or even 8W in IDLE is sick , but 9-10W full CPU is even sicker

Atoms are very efficient. When they're coupled with a GPU that comes from the mobile phone world, you get a very nice peak power consumption.
If only the laws forced them to open up their drivers...

You can get a low idle consumption with more powerful CPUs. Maybe someday we'll see well-distributed mITX boards for mainstream CPUs that can match the efficiency of this one. It's not rocket science.

Fine, now the only thing that is keeping me from buying this mobo is the missing SpeedFan support.
Is it possible to setup the case fan speed in the BIOS like this?

I have now added readings for Mageia 2 beta which uses the latest Linux 3.3.0 kernel that includes an experimental framebuffer/drm kernel driver for the CedarView GPU (I managed to find a workaround for the bug I mentioned in my previous post).
Mageia draws less power than Centos 6.2, but still more than Win7 which is unsurprising as the GPU kernel driver is still very basic and classified as experimental.
The lowest reading I managed to get with Mageia is 8.7 Watts, see article at link above for all other results.

Yes, there are similar setup options in the DN2800MT BIOS, but what do you need a fan for?
The great thing about this board is that it runs effortlessly fanless, you don't even need a case fan (unless maybe if you use a tiny case with no air vents at all).

I knew that, my plan is to install a case fan and make it run only if the temperature raises over a certain value, as a security measure.
Is it possible to configure the fan "stop temp" in the BIOS?

Ok, good job on testing this! I can't believe Intel messed this up so badly, D510MO is the best computer I ever had!

I read somewhere about someone reverse engineering the PowerVR to build a proper linux driver, can't find it again though, anyone?

So now we're back to D525MW for linux... or if someone releases a MINI-ITX with D2700 and ANOTHER GPU or a PCI x16 port! Has anyone heard anything of the like?

I have no experience with the fan settings as I don't use a fan with this board, but from looking at the BIOS settings I don't think you can set a stop temp directly. Here is a screenshot of the fan config bios page, I hope that helps you:
http://www.linuxtech.net/files/dn2800mt ... -setup.jpg

I have not heard of anything like that, but I would definitely like to see it. I believe the people interested in the "x86 with arm like power draw for router/server" thread would be as well.

In particular, I'd like to see these Cedar Trail Atoms paired with the most simplistic VGA display adapter possible, i.e. something like what comes with server motherboards. Just something usable enough to do OS installs and (hopefully rare) troubleshooting. I don't know if Supermicro is planning on doing this, but they did something similar with the Pineview Atoms, for example, the X7SPA-HF-D525. ServeTheHome reviewed a similar board and suggested about 23 Watts idle. Presumably power consumption could be further reduced with the newer Atoms (although you get a lot more functionality, e.g. dual NICs, IPMI, etc, relative to a barebones board like the DN2800MT).

In my layman's opinion, there are clearly two markets for the Atom: (1) the media/home theater PC, and (2) the low-power server/network appliance. For the former, obviously you need a competent GPU, and at least for MythTV, Ion has served this segment fairly well. For the latter, you need minimal video capabilities; I'd even be willing to give up video all together if the unit had serial output and a good IPMI/SOL implementation.

As a Linux/MythTV user, I was quite excited about the potential of Sandy Bridge's integrated GPU: capable graphics, a real processor, and low power consumption. Unfortunately, stable and comprehensive support for the GPU on Linux lagged about a year; only very recently have I finally been able to have a full-featured MythTV frontend without an NVidia chip in the mix. If this is any indication of Intel's "commitment" to Linux, it might be a while before we see decent drivers for the new Atom's GPU. I suspect the situation will be even worse because it's 3rd party tech, rather than Intel's own.

People are already selling Cedar Bridge with a dedicated GPU. I haven't seen such a mITX board though. But what would be the point? Dedicated graphics would suck so much power and I/O... why not go with one of the old IONs while you're at it (assuming you want NVIDIA, get a Zacate instead if you want ATI)? Some of them can be overclocked if you need a bit more CPU.

The GPU that ships with Cedar View is fine for servers as linuxman demonstarted.
Support should improve in the future but even with the GPU wasting power as it does without proper drivers, it's still an improvement over any other readily-available consumer mITX board.

edit: One of the new Atoms boards has two NICs by the way. And a couple have serial ports (they probably all have headers).

This:
http://www.zotac.com/index.php?page=sho ... mid=100178

I don't agree with your conclusions. Yes, CedarView Atoms are currently not very suitable for desktop or HTPC front-end use under Linux, but they are already better than anything else x86 based (apart from much slower cpus like the Via Eden or VIA C7 ULV) for headless server use (router/gateway/firewall/home server) as shown in my tests a DN2800MT based server with the current 3.3.0 kernel idles at 10.1 Watts. There is no Pineview Atom board (not even any of those based on a Netbook cpu) or any AMD Fusion E3/450 board that manages such a low power draw at similiar cpu performance and that therefore can run fanless without an expensive passive cooling case.

Also I just got an email today from one of the kernel developers working on the kernel driver for the Cedarview Atom (who is also an Intel employee) saying that the kernel GPU driver is still work in progress (it is labelled "experimental" in the kernel for a reason) and therefore should improve over the next few kernel releases.

That said, after I managed to get Mageia 2 beta working with the Cedarview kernel driver, it worked quite well, certainly usable enough as a light desktop system for browsing, email or other general desktop use (as long as there is no need for HD video playback).

Boards like the Zotac that include a nvidia GPU can be a niche solution for some cases but IMHO they are too expensive and the power draw of the discrete GPU negates any advantage of the low power draw of the Atom cpu.

Pineviews run just fine in a regular open case without any fans. And so would the Zacates which come with a good heatsink I assume.
But if you can't afford too many holes in your case (maybe you need protection against liquid spills) and want a dual-core, the DN2800MT seems to be the only suitable mITX board available to retail customers.

I beg to differ, netbook (Nxxx) Pineview cpus maybe but the desktop versions (Dxxx) cannot run safely without at least a case fan. I have an Asus AT5NM10T-I which has a large chunky heatsink, but it still gets way too hot if I let it run without a fan sitting on a table without any case at all. Zacates get even hotter (in my experience), probably due to the more powerful GPU.

"Way too hot" is subjective. Failure rates aren't.
I've got half a dozen of the cheap Intel desktop Pineviews running without any fans, mostly during business hours on weekdays. And most of them are configured to go into standy when they're not used (which typically happens several times a day). So they experience regular thermal stress unlike a 24/7 server. I have more Pineviews running fanless but they haven't been stressed as much as this bunch. I got no failures so far.
They're designed to take a lot of heat. Check Intel's technical documentation. They've designed these boards to run fanless in an open case.

Asus' heatsinks are poor. Zotac has a better one on their Zacate board but I didn't try it because they're too expensive.

edit: details

linuxman --

Thanks for your detailed idle power comparisons between the various OSes.

I have a DN2800MT being tested in the lab right now, but my power measurements differ from yours. The main difference in the gear is 4GB RAM instead of 1GB in yours, and the use of a "level V" efficiency 12VDC power brick. The latter, at full load (60W), is supposed to reach 87%, compared to 85% for level IV. Of course, at <20W, not only are both bricks less efficient, the difference is extremely small (in measurable watts).

A third major difference is the AC power meter being used. It looks like the Profitec KD-302 you use is more or less a competitor to things like the Kill-a-Watt. My AC meter is a an Extech 380803 -- said to be accurate for power measurements to within 0.9%. I get ~12W in Windows 7 idle under the same conditions as your Test 1.

One other factor: You're using 230VAC while I have 120VAC -- your PSU should run a bit more efficiently than it would at 120VAC.

Then there's that old wrench -- sample variance.

In any case, there's little doubt this board is the most energy efficient we've ever had in the lab. Typically under 14W during HD video play, 16~17W absolute max AC.

Interesting, I would guess that your 60W power brick (and possibly the additional RAM) might be the main reason for the difference, since at 10-12W it's at or below 20% load, while my 36W power brick is at 27-33% load which should mean a higher efficiency even considering the slightly lower efficiency classification.
I have no idea how much difference 120 vs. 230 V AC makes but that surely could be part of the difference too.

Do you have a 40Watt power brick to use instead (maybe the MeanWell that was mentioned in this thread)?

Did you use 1x4GB or 2x2GB? I have 2x2GB available so I could test how much power draw they would add in my tests.

What SSD did you use, the same one I used?

Did you have "OS ACPI C2 Report" enabled in the BIOS?

Can you try "Test 3" of my test series, i.e. disconnect mouse and keyboard and wait 10 minutes until Windows switches the screen into DPMS standby?
With Test 3 we would eliminate differences in keyboard/mouse power draw.


The Profitec KD302 has been measured by the most well known German computer magazine to be consistently accurate within 1-2% over a wide range of loads, so the power meter should not make a noticeable difference.