[Arm-netbook] Embedded Open Modular Architecture

Gordan Bobic gordan at bobich.net
Tue Sep 20 21:49:13 BST 2011


On 09/20/2011 08:45 PM, Luke Kenneth Casson Leighton wrote:
> On Tue, Sep 20, 2011 at 5:14 PM, Bari Ari<bari at onelabs.com>  wrote:
>> On 09/05/2011 04:49 PM, Luke Kenneth Casson Leighton wrote:
>>>
>>> for example i originally thought of doing 2x 5.0v and 4x 3.3v pins,
>>> but then i realised that, actually, that's a hell of a lot of power!
>>> it's 5 watt plus 6.6 watts, and there's no way that you can dissipate
>>> that much heat!  so, 3.3v LDO ICs being damn cheap, i figured it's ok
>>> to expect 5.0v @ 1amp and to do conversions to all power levels
>>> required, on the PCMCIA card itself (PCMCIA pins are 0.5A each).
>>>
>>>
>> Moving that much heat away from a board of this type is not a problem,
>> even without fans.
>
>   :)
>
>> Enclosures and thermal management seem to be considered 'black arts'. I
>> guess they don't teach EE's about this since it helps products die
>> around the time the warranty is up.
>
>   *grin*.  yeah my friend geoff knows his stuff - but the real issue is
> that many chinese OEMs don't!  so, i felt it safest to just say "ok,
> that's the limit guys".
>
>   funnily enough, intel's latest 1.2ghz mobile CPU would actually be ok
> (apart from the complete lack of interfaces for EOMA/PCMCIA
> compliance!  who in god's name does an embedded CPU with no bloody
> on-board interfaces!  oh wait, intel does) - the Z510 TDP is 2W.
> juuust within limits.
> http://ark.intel.com/products/35469/Intel-Atom-Processor-Z510-%28512K-Cache-1_10-GHz-400-MHz-FSB%29
>
>   hey, gosh - that's actually available on avnet and arrow.  wooow.
> there's even a price - $35.  amazing.

As of recently I am completely unconvinced by Intel's TDP numbers. N450 
is supposed to be 5.5W + 2W for the SB, so a total of 7.5W, IIRC. The 
motherboard (which also has a 2x 1Gb NIC and a CF card plugged into it) 
is drawing 22W idle, and 28W at full tilt. Sometimes at power-up it 
trips the breaker in my PoE switch which is limited to 30W/port. Even 
accounting for a relatively inefficient DC-DC PSU (PoE -> 12V DC) that 
is still way, way above what might be sensibly expected. Now, granted, 
it is possible the MoBo OEM's (Advantec) engineers are just too 
incompetent to understand power management (they are, when I queried why 
the voltage control in SpeedStep doesn't work on their MoBo they said it 
isn't implemented because it is an "overclocking feature", and/or maybe 
they got some QC reject CPUs that they have to overvolt to get them 
working reliably, or maybe they are just overvolting them to ensure 
failure shortly after the warranty period. Or maybe they just screwed up 
because they haven't got a clue what they are doing. But it is hard to 
imagine that even in that worst case scenario they managed to get the 
total TDP up to > 20W without the CPU's TDP being at least somewhat 
understated. And I have two of those MoBos so it's not a random one-off 
faulty board.

For comparison, the SheevaPlug comes in at 7W flat out, and the AC100 
comes in at 6W at full clock speed with the screen on and at full 
brightness - and that includes the AC->DC power conversion in both cases.

Gordan



More information about the arm-netbook mailing list