[Arm-netbook] uh-oh....

[Gordan Bobic]

On 12/28/2011 09:20 PM, lkcl luke wrote:

>> I believe he mentioned something like 12W with 2 Gbit
>> ethernet ports.
>>
>> For comparison, I have a couple of Atom N450 machines - the best that
>> Intel has at the moment when it comes to low-power. They are pulling 22W
>> idle, and they trip the 30W PoE Class 4 limit when everything is going
>> at full tilt. So we are talking at least 2.5x lower max power usage,
>> probably better.
>
>   ouch.  i didn't know that.  i didn't realise it was that bad.
> that's... deeply unimpressive.

To be fair to Intel, I suspect the OEM of the motherboards is partially 
at fault. The motherboard in question is this one:

http://buy.advantech.eu/AIMB-212N-S6A1E/AIMB-212N-S6A1E/model-AIMB-212N-S6A1E.htm

I know for a fact that they failed to implement PHC for stepping 
voltages along with clock speeds - I queried it and got told they didn't 
implement it because it is an "overclocking feature". No, I'm not kidding.

They also have a BIOS that allows to set different FSB speeds but 
doesn't actually do anything and always runs at default FSB clock speed.

Changing RAM timings also doesn't do anything - whatever you manually 
set, it always runs at the timings set in the SPD on the RAM chips.

There are no BIOS fixes to make the FSB and timing selection work available.

So at least the idle wattage is unduly high. But for a CPU+SB combo that 
is supposed to have a combined TDP of 9W, 30W draw, even assuming 80% 
efficiency of the PoE splitter, is still eyewateringly wrong.

FWIW, Dell Mini 1012 was also suffering from the same problem 
(unimplemented voltage stepping and possible CPU over-volting that makes 
it run so hot you cannot type on the keyboard after 2 hours of heavy 
compiling). And that is also based on the N450.

Anyway, while a part of the blame lies with the motherboard 
manufacturers, all that does is make already high CPU power consumption 
worse.

Gordan