[Arm-netbook] device tree not the answer in the ARM world [was: Re: running Debian on a Cubieboard]

[Luke Kenneth Casson Leighton]

rob i'm going to reply only to arm-netbooks, now, because the point i
wanted to make to LKML has been made.  also this is not so much
related to debian-arm so am removing them too.  i will only reply once
to david, james and robert cc'd, then remove them from future messages
as well, they can if they wish follow the discussion at
http://lists.phcomp.co.uk/pipermail/arm-netbook/ and/or subscribe to
the list.  hope that's ok with everyone.

On Thu, May 9, 2013 at 1:25 AM, Rob Landley <rob at landley.net> wrote:
> On 05/08/2013 03:19:23 AM, Luke Kenneth Casson Leighton wrote:
>>
>> On Wed, May 8, 2013 at 4:44 AM, Rob Landley <rob at landley.net> wrote:
>>
>> >>  whereas the EOMA initiative is at the complete opposite end of the
>> >> spectrum.  and products based around the EOMA standards, although
>> >> there is a cost overhead of e.g. around $6 in parts for EOMA-68, there
>> >> is a whopping great saving of 30 to 40% to the customer when compared
>> >> to other products *if* your end-user is prepared to swap / share CPU
>> >> Cards between two products.  if they share the CPU Card between three
>> >> products then the saving to them is even greater.
>
>
> It's only "whopping great" if it allows them to lower the absolute cost of
> the product. If it just buys you a 30% cost savings in a niche with an 18
> month half life of hardware depreciation, sheer inventory management can get
> you that much.
>
> The big limit of moore's law is these days you should be able to 4 megs of
> memory for a nickel, and you can't.

 i'm familiar with the costings of ICs, esp. the implications for
mass-volume.  the NREs are basically what kill you.  a 10in 28nm wafer
(bought in lots of 16) costs around maybe $5k possibly $8k each, and
you get depending on die size about.... 4,000 separate dies.  the
problem is that the NREs - the mask charges - are $2m, need to be paid
cash up-front and you get to wait 18 months in a queue.  if you didn't
pay enough at the time, and your chips run out in supplying customers,
tough luck, you get to wait another 18 months.

it's a tough business!!

 so, 1gb of 4 Hynix x8 DDR3 800mhz ICs costs you under $4 in volume.
that's very very good.

>> > In theory, Moore's Law says that buys you... 9 months?
>>
>>  and 6 months in to that 9 months you bring out the next CPU Card, and
>> the next, and the next, and the next, and the next.
>>
>>  there's a hell of a lot of history already behind the EOMA
>> initiatives.
>
>
> At what point does that history become a downside?

 my take on that is: when you're too old, too tired and too fed up to
keep on top of it and keep innovating, and have gone into
highly-optimised profit-maximising "efficiency" mode.   see the book
called "the other side of innovation" for ways to avoid that
happening.

> (We've had 20 "year of
> the Linux on the Desktop" announcements. Nobody pays any attention to new
> ones, too much crying wolf.

 luckily, i'm not focussing on "linux on the desktop" announcements.
also luckily, what linux users want is not *actually* very important
[as far as selling in mass-volume to the average end-user is
concerned].

> If I want a cheap plastic Linux system I can buy a raspberry-pi ($35) or
> pandaboard-black ($45 and the HDMI driver isn't a binary-only blob). Do
> these systems participate in your EOMA thing?

 i've spoken to the raspberry pi team: they were extremely hostile.  i
was rather taken aback at the assumptions that were made, the mistakes
that they made and the accusations that they levelled at me.  i
learned afterwards that this is the way that they behave towards quite
a lot of people.  anyway i offered them an opportunity in good faith,
and they pretty much spat in my face, so i decided not to speak to
them again.

 that having been said, the door is always open - EOMA-68 is an open
standard, after all.  i just won't put them in touch with the
factories that i work with, because their attitude could potentially
jeapordise relationships with our clients.

 the pandaboard-black ... i assume you mean the beaglebone black?  by
the $45 it has to be that one.  yes.  i've looked at that processor:
it's not enough.  people on arm-netbooks will be familiar with the
analysis process by now, but i'm quite happy to go through it with you
rob.  here's the assessment criteria:

* EOMA-68 interfaces are mandatory.  they are: RGB/TTL, SATA, USB,
Ethernet, I2C and 8 pins Digital GPIO that's 5V TTL tolerant.   the
RGB/TTL levels must be electrically compatible with an SN75LVDS83b [i
had to pick _something_ as an electrical standard and that was as good
an IC to pick as any by way of example]

* additionally, there is the general expectation that CPU Cards will
use the user-facing front plate for things like USB-OTG, Micro-SD and
HDMI output.

* the AM335x on the beaglebone has: USB, RGB/TTL, I2C, Ethernet and GPIO.

* the missing features therefore are: HDMI and SATA.  SATA can be
added using USB-to-SATA (at the cost of reliability)

* therefore, both must be added.  the HDMI IC costs say $2.00, and the
SATA costs around $1.  *but*, the SATA *also* requires a USB Hub IC
because you don't have enough USB interfaces on the AM335x.

* so in *ADDITION* to the (approx, guessing) $7 price of the 1ghz
AM335x, you must also add $2.00 for HDMI, $1 for SATA and another
$1.50 for an extra USB Hub IC.

* the $7 processor has just jumped to being a $11.50 solution.

now you compare that against say the Allwinner A10 which has just
dropped to $7 in large volume (due to the introduction of the A20) and
it is an easy financial decision to rule out the AM335x purely on
price alone.

then there is the additional disadvantage of the AM335x as a solution,
which is that because of the HDMI Converter IC, you cannot do dual
independent screens.  the additional disadvantage on top of _that_ is
that the USB-to-SATA conversion is not as good as native SATA, is
bandwidth-limited to 480mbit/s when even SATA-I is 1500mbit/sec, and
not only that but many USB-to-SATA converter ICs are known for being
unreliable.

additionally, it's complex when compared to e.g. using the A10.

all in all it's a very poor solution compared to the alternatives.

these kinds of assessments are made every time a new processor becomes
available.


> Would they benefit from it if
> they did?

 that's a very good question.  given the pricing and features of
allwinner's products, it's more a question of "will anyone ever
actually buy them *at all*"?  and this is a question which is
generally being answered with a resounding "no", not just by EOMA-68
but by *everyone*, world-wide, as you can see from this article and
the discussion here on arm-netbooks:

 http://hardware.slashdot.org/story/13/05/08/1636217/chinas-allwinner-outsold-intel-qualcomm-in-tablet-processors-in-2012


> If so, given your history, why don't they?

 there are several reasons.  firstly, these low-cost processors are
*too* low down in specification (mainly the lack of SATA).  secondly,
it's a matter of belief.  it's not like i've not given them the
opportunity: they simply don't get it.  and that's fine: i'm happy to
work with people who do believe in what is being achieved, here.
*thinks*.... those are the main ones.  yeah.

>
>> > A phone is a mass-produced consumer electronics device. Is "I can rip
>> > the
>> > guts out of my DVD player and re-use it" a commercially interesting
>> > statement?
>>
>>  you've missed the point.
>
>
> Agreed. That's why I keep asking, trying to figure out what the point is.

 ok, let me try to illustrate.  compare the following two scenarios:

 * DVD player.  with EOMA-68 CPU card.  upgrade time: 10 seconds.  a)
push button.  b) pull CPU card.  c) put new CPU card in.  d) push
until flush with case.  e) power on.

 * DVD player.  with replacement PCB.  upgrade time: several hours to
several days.  a) find an expert in electronics.  b) spend several
hours dismantling the product which was never designed to be
disassembled.  c) at some considerable risk extract PCB from morass of
wires d) replace PCB.  e) reassemble device with extreme care f) pray
like hell you didn't break anything g) swear a lot when it goes up in
smoke h) go out and buy a replacement DVD player.

is that clear at all as to how easy it is to upgrade
EOMA-68-compatible products to the latest-and-greatest firmware and
latest-and-greatest CPU?

 it's so simple that, as a very good friend of mine puts it: "i can't
make this complicated enough for you to be able to understand it".

 push button.  eject CPU Card.  insert upgrade CPU Card.  power on.

 it's THAT simple.


>
>> EOMA-68 CPU Cards are separately-sold
>> mass-volume *interchangeable* products, i.e. being packaged in legacy
>> PCMCIA housings they have the exact same advantages of PCMCIA except
>> now it's the *CPU* that's interchangeable between products.
>
>
> More mass-volume than phones?

 i'm not interested in the phone market - it's too complex, and too
specialist.  EOMA-68 is targetted at the "pretty much everything else"
market, from DVD players and Media Centres, to LCD Monitors, HDTVs,
budget Desktop PCs, Laptops, tablets, NAS Boxes, Smart Routers, SoHo
servers, Games Consoles, FreedomBoxes - all the sectors (with the
exception of the freedombox) which are likely to sell in numbers which
just completely blow your mind.

> I'm trying to think of the last time I got a new nebook with legacy PCMCIA
> in it. It's been more than 5 years...

 yes.  fortunately, PCMCIA is still being used in the Satellite
"Conditional Access Module" market, which means that it's still being
produced.  it's been an absolute bitch tracking down suppliers though,
but we found some.

>
>>  nobody in their right mind swaps the DVD electronics, they just buy
>> another DVD player.  including the mechanical part and the built-in
>> PSU, and the GPL-violating software running on it.
>
>
> Yes, that was sort of my point. What's _different_? And how is this not
> http://xkcd.com/927 ?

 love it :)

 ok.  let's take a look at the competing standards.  i won't reiterate
what's on this page, but this page's contents *is* the answer:
 http://elinux.org/Embedded_Open_Modular_Architecture#Alternative_Standards

 can you see how i go to some lengths to illustrate that EOMA-68
learns from the lessons and mistakes of those standards?

 EOMA-68 is targetted at *MASS VOLUME* products, for *END USERS* to be
able to upgrade their products.

 the Q-Seven Standard is targetted at *FACTORY INSTALLATIONS ONLY*.

 what people who write these standards do not fully understand is that
it is *essential* not to have anything that is optional.  i cannot
express enough how essential that is.  there are not enough words to
describe strongly enough how essential it is.

 there is only one other modern computing standard that does not have
optional-itis: it's COM-Express.  unfortunately, COM-Express is
primarily targetted at the embedded / industrial market, and as such
it is hugely more expensive to use.

 EOMA-68 is targetted *right* at the start at the mass-volume market
*NOT* at the industrial market.

 does that help make it clear what the differences are?

> You had five minutes worth of my interest, but that was not a five minute
> list of links.

 indeed.

>> http://www.c2mtl.com/eye50/ideas/the-rhombus-tech-eoma-68-initiative/
>> http://rhombus-tech.net/articles/eoma68_in_education/
>
>
> Ok, obvious from those first two links that rhombus-tech is behind it.

 yes - that's me.

> I don't think I'm the target audience here.

 that's fine.  there are plenty other people who are.

l.