The bottom line is: You've told us you have problem X, and you're getting suggestions based around that. If you really mean "Someday I will have problem Y and I want to be able to solve that problem without ever having to change the hardware" then the suggestions might be different.
What you have to ask yourself is:
- Will customers with old units already fielded be interested in whatever new bells and whistles you can come up with in software upgrades, or are they just paying for hardware they'll never use?
- Are you sure that you're *ever* going to need any functionality of Linux, or the mini-ITX solution? Is it likely that upgrades to use such functionality will correspond to major product line remakes anyway?
If your customers are willing to pay $400 for a unit that does X,Y,Z functions (ONLY), then what sounds more sensible: spend $50 build cost per unit on an optimized solution and make $350 profit, or spend $150 per unit build cost, make $250 profit, and hope that future software upgrades will earn you >>$100 net?
$60 - motherboard $15 - PSU $20 - RAM $20 - bootable media
$115 + housing.
Your route also has complications with EMI (it's a big, fast circuit), and appallingly high power consumption for the described application.
Remember that mini-ITX hardware follows *PC* life cycles. In 6-9 months you will not be able to buy absolutely identical hardware. In
12-15 months you will probably need minor mechanical rework to fit whatever is on the market. In 24 months you most likely will not be able to buy one single part that is in today's design.I'm not saying Tern's products - or indeed ANYONE's products - are the answer - I think a cheap all-in-one microcontroller would be way more than you need, and it would have a total BOM of $40 or so including housing - but I hate seeing a really suboptimal solution being used when there is really no justification beyond a vague assertion that "it might be useful someday".