[time-nuts] RPi/ beagle bone-like computer without video
jimlux at earthlink.net
Fri Dec 2 12:13:28 EST 2016
On 12/2/16 8:51 AM, Attila Kinali wrote:
> On Fri, 2 Dec 2016 08:05:17 -0800
> jimlux <jimlux at earthlink.net> wrote:
>>> I'm measuring 0.350A with max cpu usage on all cores and the following
>> @ 5V, right, so 1.75W going full out. That's a useful number to know.
> I once did an board with an i.m233 (arm9 at 400MHz) and it did less
> than 0.5W IIRC. But I think this is about the lower you can do without
> tweaking the OS.
>>> Apart from that I read somewhere that the beaglebones have terrible EM
>>> characteristics. But I can't find my source so take it with a grain of
>>> salt :-)
> Yes. The BBB is badly designed in this regard... The RPI is better there.
>> I must say, there are more of these things available in almost too wide
>> a variety.. You could spend days going through all the datasheets and
>> websites - I suspect that they ALL have about the same power consumption
>> for a given amount of processing horsepower - same feature size on the
>> die, after all - so it's more about peripherals and ease of use
> There are way too many, IMHO. And a lot of them are not usefull for
> a lot of stuff or have very bad support. The Odroid are a prime example
> of this. There is a "community" around them, yet getting them to do
> anything usefull is a major pain. There is a handfull of companies
> I know of, who do provide good support and those are the ones I am
> usually sticking to (unless I have special needs).
>> And, it's more likely that idiosyncracies in the distros have been
>> identified and it's more likely that the software will run on them after
>> its built.
> It's actually better to go with a company who is invested in giving
> you a working board than using something popular. Especially one that
> cares to push all its patches upstream.
> Beside the mentioned Toradex, and Aries Embedded, there is also Olimex
> which is known for it's wide variety of boards with good support.
> Depending on your exact requirements, I would probably go for one of
> the i.mx233 boards (the imx233 nano is quite neat) or A10 or A33 board.
> Especially the i.mx233 is nice as it has an on-chip Li-poly charger/controller.
> All you need to do is to supply it with 5V and it does the rest.
> A note of warning: a lot of the boards from Olimex have not enough ground
> pins for the high speed singals they provide. If you are transfering data
> with high-speed (several 10MHz) over the headerpin connectors, you will need
> to add some additional ground connections.
> What are the exact requirements you have? How much computational power
> do you need? How do you interface the sensors? How many boards will
> you need? Is it out of question to build your own processor board using
> one of the ARM9's in QFP? What is your budget?
We're processing several thousand samples, received over a serial port
or USB in a few seconds. The algorithm (in Matlab, hence the need for
Linux) grinds for around 30 seconds to produce the output.
we're not sensitive on the "board cost" - labor to design a board is
expensive, so a board that has low power, and the right connectors, so
it's <1 day to make cables, etc. is a better deal than several weeks to
design a board and spin it, etc.
> The reason why I'm asking the last two questions is, it is often more efficient
> to do your own CPU board if you have to design a PCB anyways for the sensors,
> need more than 10-20 boards and you can live with one of the "small" ARM9's
> that come in QFP packages (like the i.mx233 or AM1705).
That would come later, and be "someone else's problem" - We do the proof
of concept, "demonstrate that it works in a relevant environment", and
then it goes from there.
More information about the time-nuts