Intel DN2800MT hands on...Single digit world?

[tramall]

Hello

Finally got my hands on Intel's newest atom board. So , let's play.

Setup: Intel DN2800MT , 2GB RAM , Kingston mSATA SSD , 12v AC/DC 4.16A brick(not the best one , some random generic stuff).
Software: Windows 7 Pro , K-lite codec pack 8.4

Power usage:

Idle: 7W (stays most of the time)
1080p : 8W-9W .
Max CPU(LinX Problem Size 10000 , Memory to use 772MB) : 10-11W .

I do not know if my power meter is broken or not that sensitive. Just writing what I see.

My impressions:
Board is nicely finished , good quality. It is still atom (it takes 79 seconds to do 1M Pi sample) , overall performance is on par with Atom D525 , sometimes a bit faster. It is fanless , and by that ,I mean Fanless. Used to own a Zotac ION which had to have a lot of airflow in the case (mainly because of the GPU). The board was tested in small case (200 x 200 x 45mm) and did not exceed 60C (cpu temp).

Problems: It seems that drivers are not optimized (graphics) , and testing software (speedfan and coretemp) could not properly recognize this board (coretemp 1.03 stated that cpu has 11,2Ghz , 133 x 84 , older version displayed it as a engineering sample) . Speedtemp said that cpu was 111C hot.

Overall: For me, it is a D945GSEJT with faster cpu and broadcom hd chip . Nice try Intel, but still waiting for Silvermount

Thanks to the http://www.get-solid.pl guys for providing the sample board . (It is still not available in Poland)

[HFat]

Your power consumption measurrment is consistent with the other I've seen. Finally some progress.

The previous generation of Atoms also worked well fanless but didn't have digital video output (except for the single-core D945GSEJT and only with a limited resolution).
The trouble with this and the other new Atoms is the graphics driver which is not available for 64bit Windows and only as a userspace binary in Linux.

If you're so inclinded, it would be interesting if you could check if the graphics driver included in this works well: http://repo.meego.com/MeeGo/updates/1.2 ... cedartrail

[onlinespending]

I definitely want to get this board. It says it takes up to 4GB of RAM. Do you know if it can take one 4GB sodimm, or would you need to use 2x2GB? Thanks.

UPDATE: I think the answer is Yes. From the product guide:

Code: Select all

Support for up to 4 GB of system memory on a single SO-DIMM (or
two 2 GB SO-DIMMs)

[linuxman]

If you're so inclinded, it would be interesting if you could check if the graphics driver included in this works well: http://repo.meego.com/MeeGo/updates/1.2 ... cedartrail

Thanks for the link, I should be getting a DN2800MT early next week and will definitely be testing this Meego ISO and report back.

[HFat]

I think it's likely it can take more than 4G actually.
People have claimed to use 8G with other boards Intel says support 4G. Intel probably stated that because of the lack of 64bit drivers.

[onlinespending]

This board is tough to find in stock. Just bought one from Buy.com through their eBay store. They have a few more in stock, FYI. I'll report back once I get it and set it up.

[linuxman]

I have this board now, but I'm struggling to accieve single digit power consumption with Centos 6.2 (headless, GPU disabled in BIOS). I'm seeing 12.5 Watts idle which is more than I had with Centos 6.2 on the old D945GSEJT (10.5 Watts idle with the same power brick and a PCI 10/100 NIC card). I briefly tried the Meego ISO mentioned earlier and that one was around 7.5-8 Watts, but Meego is next to useless for me.

Has anyone else tried this board with Linux so far?

[HFat]

So you're apparently getting the benefit over the D945GSEJT other people have measured, but only with the driver Intel put in MeeGo. Have you tried to run MeeGo on the D945GSEJT to see what power consumption you get in that case?
What does disabling the GPU in the BIOS do to power consumption? Maybe this feature doesn't work properly.

[linuxman]

So you're apparently getting the benefit over the D945GSEJT other people have measured, but only with the driver Intel put in MeeGo. Have you tried to run MeeGo on the D945GSEJT to see what power consumption you get in that case?

There is no need to try Meego on the D945GSEJT, as the D945GSEJT is fully supported by any standard Linux distro.
In fact the 10.5 Watts including a PCI NIC card I get with the D945GSEJT and Centos 6.2 is excellent for that board (and probably the minimum possible with that board), previous attempts with older distros were around the 12 Watt mark.

The whole point why I bought the DN2800MT was to accieve lower Watts than the D945GSEJT, while at the same time benefiting from the extra performance of the dual core cpu.

Also I don't really care about the GPU as the D945GSEJT was running as a headless server and the DN2800MT is meant to replace it for the same task. The problem appears to be though that even when running headless there needs to be a GPU driver to put the GPU into it's lowest power state, disabling it in the BIOS or not attaching a monitor doesn't seem to be enough.
Even with the D945GSEJT I was getting the 10.5 Watts only when booting in frame buffer mode (GPU gets initialised by the kernel driver), pure text mode (with no GPU initialisation) increased the consumption by 1 Watt (IIRC).

What does disabling the GPU in the BIOS do to power consumption? Maybe this feature doesn't work properly.

I agree it doesn't seem to be working properly, it does disable the VGA/hdmi output, but it doesn't appear to power off the GPU. I need to remeasure to get the exact Watt figures, but there was very little difference when disabling the GPU in BIOS when I checked earlier (BTW I'm obviously running the latest BIOS from the Intel support web site).

I haven't given up yet, if the hardware can do 7 Watts idle (as proven with Windows) then there must be a way to accieve this with Linux, too!

Edit to add: I'm currently installing Mageia 2 beta 1 on the DN2800MT, I hope since this distro release is bleeding edge (kernel 3.3 rc6) that it will have better support for Cedar Trail.
If Mageia fails to deliver I'll try Fedora 17/Rawhide.

[HFat]

There is no need to try Meego on the D945GSEJT, as the D945GSEJT is fully supported by any standard Linux distro.

Sure but the point was to rule out discrepancies other than the drivers which could explain some of the differences in power consumption.

The whole point why I bought the DN2800MT was to accieve lower Watts than the D945GSEJT, while at the same time benefiting from the extra performance of the dual core cpu.

You also get PCIe and a couple of other modern features.

Even with the D945GSEJT I was getting the 10.5 Watts only when booting in frame buffer mode (GPU gets initialised by the kernel driver), pure text mode (with no GPU initialisation) increased the consumption by 1 Watt (IIRC).

interesting

I haven't given up yet, if the hardware can do 7 Watts idle (as proven with Windows) then there must be a way to accieve this with Linux, too!

Well, you achieved it (or close enough).
You obviously need the code Intel put into the MeeGo kernel and perhaps also the closed source user-space binary even though you're running the board headless. Until someone reverse-engineers enough of this closed source crap, that is.

[linuxman]

You obviously need the code Intel put into the MeeGo kernel and perhaps also the closed source user-space binary even though you're running the board headless. Until someone reverse-engineers enough of this closed source crap, that is.

Yes, the GMA3600 code that's in the Meego CedarTrail ISO is also in the latest kernel (3.3), that's why I'm trying the Mageia beta which uses the latest kernel. Still 15 minutes to go for the installation to complete.

[linuxman]

Hmmm, Mageia 2 beta 1 disables screen output (the monitor goes into standby) as soon as the kernel activates the frame buffer during boot... Obviously the kernel doesn't like the GMA 3600 for some reason.

[linuxman]

I tried Fedora 17 alpha too, but since it doesn't come with the 3.3rc kernel it doesn't include the code for the GMA3600 yet, so it boots up fine and starts X but only in VESA mode (idle power usage is 13-14 Watts).

I think I will wait for the final release of kernel 3.3 and/or Mageia 2 beta 2, right now I don't have too much time to do loads more experimentation.

[HFat]

The regression with the newer kernel means people are working on it, which is good. Confirmed by a quick Google:
https://bugs.launchpad.net/ubuntu/+sour ... bug/944929
What power consumption we'd get without the closed source blob remains to be seen.

edit: corrected link

[batou]

just wondering: did anyone get the chance of trying to underclock/undervolt the 2800mt intel board, or their smaller-number sibilings (like the hudson, the 2500 looks ...affordable)?

i'd be curious to know - with a proper use of the bios, setfsb or similar tools - how much the Watts could go down, into the Arm territory...

[linuxman]

just wondering: did anyone get the chance of trying to underclock/undervolt the 2800mt intel board, or their smaller-number sibilings (like the hudson, the 2500 looks ...affordable)?

i'd be curious to know - with a proper use of the bios, setfsb or similar tools - how much the Watts could go down, into the Arm territory...

How would you suggest doing that? The BIOS has no options with regards to clocking or voltage.

It has some options related to C-states but one of them is enabled by default and enabling the second one didn't make any noticeable difference when I tried it.

That said the cpu obviously has Enhanced Intel SpeedStep Technology, so it already idles at at a very low voltage and at 798MHz and only ramps up under load.

As far as I can see it, the problem isn't the hardware, which IMHO is very good and very capable of low power consumption, it's the GPU drivers (or lack of them) especially for Linux (but also for 64 bit Windows), and this issue is made worse by the fact that the GPU drivers are needed to power down the GPU, by default the GPU appears to be running in full power mode.

[batou]

One possibility could be using this software (that I never personally tried) http://sourceforge.net/projects/lfsb/

anyway, since the cpu scaling is working well, I agree that there's not much (if at all) interest in trying.

I think that the driver (or at the very least some utility to fix the gpu related power issue) will come out, since intel is usually quite supportive, but we have to wait for that...

If I understood correctly what has been said in this thread, in addition to the meego driver, there's the possibility that a standard-kernel driver included in the latest kernel would give you back the 3 watts;
did you try to recompile the vanilla sources from kernel.org, without waiting for a distribution-sanctioned kernel package?

damn, maybe for me it's better to look for used z5xx netbooks, but it seems that there aren't many of them around anymore...

[HFat]

I don't know. It's not an Intel GPU and they have made discouraging noises. But because these Atoms are going to be a lot more common than the Zxxx, people will most likely work on getting halfway decent support in the kernel much sooner.

[linuxman]

If I understood correctly what has been said in this thread, in addition to the meego driver, there's the possibility that a standard-kernel driver included in the latest kernel would give you back the 3 watts;
did you try to recompile the vanilla sources from kernel.org, without waiting for a distribution-sanctioned kernel package?

Yes the GMA3600 related code that's in 3.3.rc6 apparently has a bug that got fixed shortly after rc6 was released and will be included soon in Mageia (the kernel packager of Mageia confirmed this). I haven't tried building the latest kernel snapshot myself since I currently don't have the time for that.
The release version of Linux kernel 3.3 should be out quite soon and will then be available for Mageia 2 shortly after, too.
I'm using Mageia here because AFAIK currently among all traditional Linux distros it's the distro with the most up-to-date kernel package (Meego is too limited to be considered a proper general purpose distro).

I don't know if the GMA3600 frame buffer driver in kernel 3.3 will give us back the 3 watts, but I really hope so!

damn, maybe for me it's better to look for used z5xx netbooks, but it seems that there aren't many of them around anymore...

Why? You would have similar problems (AFAIK z5xx Atoms only have limited support in specific distro versions) but on outdated hardware that won't get much developer attention any more. Like Hfat said, Cedar Trail Atoms will surely be a lot more popular than the Z5xx ever was and therefore there should be more incentive for developers to get them working properly.

You need to remember this is bleeding edge hardware right now, it's not even available everywhere yet.

[batou]

good to know you use mageia now, since if you didn't and by chance you still had to wrestle with centos or fedora (I'm still at 16, I tried the 17 but it's still too raw) I could have been tempted to try to build a kernel package over the weekend with the very latest commit available in the git repo (not sure, it's been at least 10 years since I recompiled a kernel, and it would have been my first homebrew rpms ), so you being with mageia actually saves me a lot of work

I vaguely recalled an already existing and working closed source poulsbo x11 driver for the z5/6s, but I see that Intel has long put it in the freezer; appropriate, for dead meat.

I'll quietly stay tuned waiting for good news; till then, thanks a lot for the clarifications

[leem]

Will Intel develop XP drivers for this board?

[linuxman]

I guess this is the kernel patch we are waiting for to be integrated into the distro kernel packages:
http://git.kernel.org/?p=linux/kernel/g ... 29b8404ff2

[linuxman]

Big question to the OP, if he/she still reads this thread:

How exactly did you get 7 Watts idle with Win 7, I mean what hardware/peripherals/cables were connected to the DN2800MT?

I have now installed a clean copy of Win7 SP1 Home with all the latest drivers (GPU, audio, chipset, LAN) from the Intel DN2800MT driver download page because I wanted to have some reference power readings to compare my Linux power readings to on my specific setup.

Hardware:
DN2800MT in mini-box M350 case
1x Samsung 1GB DDR3 8500 (1066Mhz) SoDIMM (I chose to go with the absolute minimum for this test on purpose)
1x Intel 311 mSATA SSD 20GB (Intel claims this SSD only uses 150mW when active, so next to nothing)
ELV 87586 36VA power brick set to 12V
Watt meter: Profitec KD-302 (known to be very precise even at very low Wattage, usable range: 0.2-3600W)

When WIn7 is sitting idle at the desktop with 1 USB keyboard and 1 USB mouse connected, 1 Gige cable connected with idle link and a 17" LCD (1280x1024) connected via VGA I read 10.1 Watts (nowhere near 7 Watts!)

If I disconnect the USB mouse and keyboard (with integrated mini USB hub), the power consumption drops to 9.7 Watts.

Once Win7 switches the monitor into DPMS standby after about 10 minutes idle, the power consumption drops to 9.1 Watts.

If I now detach the ethernet cable, the power consumption drops to 7.7 Watts!

I can't get the power consumption any lower than this with Win7. What made a big difference was disconnecting the ethernet cable, so I guess when the OP did his tests he also didn't have an ethernet cable connected.

In view of all this the Linux power readings weren't that bad at all, yes they were still higher but not 5-6 Watts higher as I was thinking based on the OP's power readings (7 Watts Win7 vs. my 12-13 Watts Linux).

Now once the Mageia 3.3 kernel package is available I will do the same tests with Mageia Linux.

[linuxman]

One additional note:
when enabling "OS ACPI C3 Report" and "PCIe ASPM Support" in the BIOS (both are disabled by default) the above readings drop by another 0.5/0.6 Watts, so the lowest reading with Win7 idle, keyboard and mouse detached, ethernet detached and monitor in DPMS standby is 7.1 Watts and Win7 idle with monitor on, keyboard and mouse attached but ethernet still detached is 8.2 Watts.

[HFat]

Can you determine your AC/DC converter's efficency, for instance with a known load?

It's good of you to have tested this but it was obvious some other factor was explaining part of the difference (like the converter or the meter). The N2800's TDP being 6.5W, the GPU couldn't possibly be consuming an extra 5-6W over its idle power consumption. My guess would be that the GPU driver can account for 2.5W at most at idle (likely less) before inefficiencies are taken into account (they would add to this difference).

[linuxman]

Can you determine your AC/DC converter's efficency, for instance with a known load?

I'm not an electronics expert so I'm not sure what a good known load could be, would a 12V 10W incandescent light bulb be ok?

with regards to the ELV power brick, I have 3 different 12V AC/DC power bricks (one from FSP, one from LC-Power and this ELV one) and in the past I compared them all and this ELV one was the most efficient one when drawing around 10W, so I would think it's relatively efficient.
Also the Profitec KD-302 watt meter was top rated during a comparison test by German c't' magazine a few years ago because it's very precise so I do think my measurements are reasonably accurate.

[HFat]

Yeah, your measurements are the best we got. But if you compare them to other people's readings, you take their errors on board.

I'm no expert either but I suppose it would be best if you could measure another board whose power consumption is known. But a 12VDC 10W incandescent light-bulb should be better than nothing. If your reading is way off, you'll know I was wrong.

[tramall]

Just to make things clear:

During testing , Lan cable was unplugged (I have only wifi network).
My powermeter is FHT-9999 , http://www.made-in-china.com/showroom/a ... 9999-.html
Power brick was Starmen TCS050120
Current was 230V AC

Rest was, as stated in the first post.

Keyboard and mouse were plugged in and VGA ELO touchscreen (touchscreen was off) monitor (19 inch 1280 x 1024) was connected.
My intention was not to fool anybody or provide false results . Just wrote, what was on the powermeter's screen.

I repeat it once again , If You want to break efficiency records I would suggest Mean Well's GS series power bricks . We use GS60 and they are excellent albeit expensive. I spoke to Mean well's rep , and he confirmed that this line of bricks is designed for 24/7 use (digital signage, vending machines, POS ) hence quality of the components is bit better. For this board I would use GS40 : http://www.meanwell.com/search/gs40a/gs40-spec.pdf.

[HFat]

My intention was not to fool anybody or provide false results .

Of course, and I thank you for providing your measurements which are useful. First measurements always are.

I spoke to Mean well's rep , and he confirmed that this line of bricks is designed for 24/7 use (digital signage, vending machines, POS ) hence quality of the components is bit better. For this board I would use GS40 : http://www.meanwell.com/search/gs40a/gs40-spec.pdf.

Maybe the rep could send a review sample to SPCR and/or other credible reviewers then. That would help small customers choose their products (if they're good).

[Vicotnik]

I repeat it once again , If You want to break efficiency records I would suggest Mean Well's GS series power bricks . We use GS60 and they are excellent albeit expensive. I spoke to Mean well's rep , and he confirmed that this line of bricks is designed for 24/7 use (digital signage, vending machines, POS ) hence quality of the components is bit better. For this board I would use GS40 : http://www.meanwell.com/search/gs40a/gs40-spec.pdf.

Some of the Mean Well adapters are available through eBay, but shipping outside US is very expensive. If anyone know a better way for us in Europe to get hold of one, please let me know.