ARM 11

[SaturnsVoid]

Is the ARM Processor a Dual Core?

[Svartalf]

No.

[scologic]

it's the cheaper cousin of the A8 Cortex - which would be a fantastic processor to have instead.

[jamesh]

Quote from scologic on September 14, 2011, 21:11
it's the cheaper cousin of the A8 Cortex - which would be a fantastic processor to have instead.


Indeed. Except there isn't a A8 with a Broadcom GPU, which means the price would be too high. The price Raspi are paying for this SoC is very very low, for reasons explained elsewhere.

[tangobravo]

From Gert's recent Q and A I see you managed to add the ARM 11 to a Videocore in a month with a small team (great work btw). I remember reading the ARM is very much a minority of the chip area too...so perhaps a similar job could add a Cortex (have to say A9 now, with NEON please...) without too much hassle. No doubt ARM's licensing fees are much higher for those but I think it would make the general computational performance significantly better.

[Svartalf]

Quote from tangobravo on September 15, 2011, 17:41
From Gert's recent Q and A I see you managed to add the ARM 11 to a Videocore in a month with a small team (great work btw). I remember reading the ARM is very much a minority of the chip area too...so perhaps a similar job could add a Cortex (have to say A9 now, with NEON please...) without too much hassle. No doubt ARM's licensing fees are much higher for those but I think it would make the general computational performance significantly better.

They've got a planned chip that does that very thing- it's just that Broadcom can't sell this chip cheaply enough to make the R-Pi pricepoints, even with the R-Pi foundation's discounts they're getting. (There's a reason I wished they'd open up the Videocore IV stuff a bit so we could have FOSS drivers and give Broadcom an advantage in the Android marketspace. )

[tangobravo]

I've just come across that chip in another thread - if we're talking BeagleBoard-type pricing then that's too much. For another $10 I'd definitely consider it worth it. I fully support the choices made for the A/B devices though, but I hope it's not too many years before something similar but A9-powered comes in at under $40-50.

[Chris]

Quote from tangobravo on September 15, 2011, 17:58
I've just come across that chip in another thread - if we're talking BeagleBoard-type pricing then that's too much. For another $10 I'd definitely consider it worth it. I fully support the choices made for the A/B devices though, but I hope it's not too many years before something similar but A9-powered comes in at under $40-50.

Even if it was £40 to £50 I would buy it, some GAMES cost as much as that, and the cheapest android smartphone is £100.

[obarthelemy]

in the other chip a drop-in replacement for the current one ?
if it is, maybe the foundation could do a run with that, and maybe expensive 512 MB RAM. If it's not, I think i's too far outside their purview to ask them to design a board just for upmarket hackers.

[liz]

It's not a question of just dropping a different chip in, unfortunately - and Raspberry Pi isn't in the business of fabbing chips. Hopefully, we'll see new generations of SoCs come along, but for now we're running with the 2835 because…you know…it exists.

[gsh]

Hey yeah, we could make a new chip, anyone got $15M lying around?

[Svartalf]

Quote from gsh on September 16, 2011, 17:23
Hey yeah, we could make a new chip, anyone got $15M lying around?

SNERK!

If I had that sort of cash lying about burning a hole in my pocket, I'd be trying to bankroll a new FOSS GPU core for the embedded SoC space.

[stuporhero]

Heh - as someone trying to peddle with OS development I'm glad there's just the one core... for now B-)

[jamesh]

Quote from Svartalf on September 16, 2011, 17:24
Quote from gsh on September 16, 2011, 17:23
Hey yeah, we could make a new chip, anyone got $15M lying around?

SNERK!

If I had that sort of cash lying about burning a hole in my pocket, I'd be trying to bankroll a new FOSS GPU core for the embedded SoC space.

Videocore cost considerably more than $15M....perhaps 10 times if not more although I don't know the exact figure. You don't get much for 15M in chip design nowadays. It costs $1M just to tape out a Videocore - if you need (and you will need) a couple of extra revisions to iron out the bugs, that's $3M before you have done ANY design or software work. The Videocore has gone through version 1 to 4 and each one had a number of revisions, to get where we are now.

On a 100M transistor chip (not saying that what Videocore is - just a example figure) you need a big team of people for design, layout etc, and a very big and very smart software team to write all the GPU code.

Let's say it costs $100M to get your chip done and ready for market. Are you going to give away all the IP and software for FOSS?

Given that, the price of a Raspi is pretty low I think!

And the A8 Cortex + Videocore IV chip mentioned above - not yet available.

[Andre_P]

For those who have not actually ever produced a chip before I'd say the absolute minimum spend would be approx $80,000 for a V simple SoC . You would get a about 20 devices.
Here is a very rough outline of the stuff that goes on for the Software Only guys and gals here (for those who are in the game huge swathes of stuff have been omitted for brevity, please do not shoot me).
Writing your design docs.
Writing your design (including Design For Test capability)
Compiliation, Simulation/Testing of your design (Assuming you can get tools but there are ways ! )
Synthesizing your design into a 'target library', The Target Library represents what the logic gates will look like as transistors on the kind of silicon you want to use. The tools are V expensive and these days you sometimes have to pay for the library'.
Placing and Routing the generated logic, ie taking the transistors and laying them out so that the design times, again tools are V expensive and this is generally farmed out to a 3rd party if you are a small company.
In parallel to all of this you are generating your Test Program so you can test the devices you have made, again this can be farmed out, especially if you have no knowledge of the tester, this uses the Design for Test stuff you put into your code.
Once you are out of Place and Route you need to make sure that the design still works and importantly 'times', ie it can go fast enough when it's hot with low voltage and curiously not go tooo fast when it's very cold and lot's of voltage (look up Hold Violations).
Once all of that has been shown to be ok you send it off in a process called 'Tape Out'. Designs used to be sent out as Magnetic Tape Reels to the Fabrication Plant.
At the Fabrication plant comes the creation of the individual masks for each layer, this is effectively shared for a Multi Project Wafer run, ie lots of people are trying out their design so effectively sharing the costs. It's a lot more expensive for a single design wafer of silicon.
Once the wafer has been made you then apply your test program to each of the 'chips' or as they are known 'die' to sort out what ones you want to package.
Then comes your packaging which means encapsulating and bonding of IO to the pins of the encapsulation.

Then you hope and pray you don't have sand when it comes back to you !

And just to show that Men can multi task while writing that I've taught a 6yr old to draw a face (big fat bum at the top smaller fat bum at the bottom), destressed a partner and got organised to attend a wedding reception ! Who says we Engineers can't get the job done to a time line !