Re: Raspberry Pi and Lua

[Jay Carlson <nop@...>]

Disclaimer: the last hardware development project I was paid to work
on was twenty years ago. I don't have to make Q10,000 BOM decisions.
If I'm incredibly lucky, a manufacturer is interested in talking, but
usually I just get to pick which piece of consumer hardware I'm going
to repurpose. Lua shows up here because my first goal on any new
platform is to get the hell out of C. On the VTech Helio port I went
with Tcl. Since the Agenda VR3, Lua has been my weapon of choice.

On Thu, Mar 8, 2012 at 10:10 PM, Alex Bradbury <asb@asbradbury.org> wrote:
> On 8 March 2012 20:16, Jay Carlson <nop@nop.com> wrote:
>> It will be interesting to see if any clones become available. If Broadcom is not selling the chip on the open market, they won't.
>
> I think it would be more interesting to see devices like the Raspberry
> Pi and with a similar price point but using SoCs from other vendors.
> Differentiating factors as you suggest could be availability of full
> documentation and possibility to buy the chip in small quantities.

To be clear, I meant the existence or not of clones will be
illuminating to outsiders as to the nature of the relationship the
Raspberry Pi project had with Broadcom. But I take your point.
Indirectly, there is a more general pressing question of how close any
consumer embedded manufacturer must be to the upstream silicon design
house in order to ship sophisticated products at competitive prices.
It's possible we're headed for *all* boutique packages if not silicon
as a result of an intersection of physics, capital, CAD, and
patent/copyright/trade-secret law. I'm beating up on RPi in particular
here but I don't know how arms-length TI is from the Beagle-stuff
either. Building a system out of components from different vendors
tends to increase modularity as the module boundaries get documented.
Vertical integration[1] is a little worrying.

The biggest differentiator for the RPi SoC *in the RPi* is the GPU;
some of the stuff the mentioned datasheet describes is pretty common
among IP blocks for ARM.

Let's look at another vendor. Realtek appears to be shipping a zillion
RTL8169s to lots of small manufacturers; the chip seems like it has
most of the other important features besides GPU for the RPi. It's
even in a non-BGA package, not that I can solder that pitch of SMD
either. I forget whether it has I2S or not, but I believe it does have
a PCIe lane which can make up for a lot of other failings--in the
usual application of the chip the lane goes to the
802.11b/g/n/whatever transceiver. DDR2 RAM will probably be back to
BGA though....

I'd send you to the datasheet but gosh, it's not available either.[2]
Sigh. At least the Linux port is pretty complete, and the Realtek dev
kit makes it possible to cut&paste the reference schematic and bring
up the OS without much software knowledge. Which is why there are a
lot of little manufacturers cranking them out. As a coincidence this
design space was also Broadcom's to lose or abandon, although I don't
know how market share looks today.

As far as the GPU goes, everything seems like it sucks in the embedded
market anyway. I found
https://wiki.linaro.org/ChristianReis?action=AttachFile&do=view&target=embedded-gpu-lca-2011.pdf
to be illuminating. (If patents really are the problem, everybody
would be better off getting locked into a suicide pact via GPL3--that
way the first design house to open up specs couldn't be sued by their
competitors, only by patent trolls.)

Jay
[1]: And sometimes *literal* vertical integration, as in this case. :-)
[2]: Well, the datasheet isn't *officially* available. I don't read
enough Chinese to navigate the sites where it appears to be leaked,
which is perhaps just as well.