[Arm-netbook] xilinx zynq-7000 800mhz ARM Dual-Core Cortex A9

[Luke Kenneth Casson Leighton]

On Sat, Aug 20, 2011 at 1:21 PM, Gordan Bobic <gordan at bobich.net> wrote:
> The other thing to consider here is whether we actually need 3D acceleration
> on the ARM GPU at the moment.

 well... here's the nice thing: i looked again at the OGP, and they
are using the Spartan 3 4000.  that's equivalent to 4m gates, but is
listed as only 65k logic cells.  the z-7030 is listed as 125k logic
cells, they say it's equivalent to 1.9m gates.

 so... that _actually_ means that yes, the CPU could itself *be* the OGP card!

> If you want to extend to a bit of entertainment use, then some DSP to get HD
> video playback is nice to have

 ah no, you can use the FPGA+CPU for that!  the expensive part is the
CABAC decoding (which, stupidly, cannot be parallelised), that would
be handled by the FPGA: you then hand the data over to the NEONs,
which now have a hell of a lot less work to do and could thus more
than likely get you up to 1080p.  we know that NEON can handle 720p
(complete decode) in dual-core cortex A9, so i reckon it's achievable.

> but for the large chunk of that use-case you
> are at Adobe's mercy WRT getting libflash.so ported.

 ah no.. you can just use the ARM CPU for that!  then when you have
access to the de-obvuscated data, deal with it from there.

> So opengraphics are actually a pretty decent idea, but in the ideal world
> the ARM coupled GPU should be optimized for minimum power (Tegra2 is pulling
> about 3W flat out, so with a basic 2D GPU you shouldn't be aiming for more
> than 2.5W at most). Whether opengraphics can achieve that - I have no idea.
> And then on top of that, assuming it is viable in terms of power draw, there
> needs to be an ARM licencee that is willing to go down this route for the
> GPU of their SoC - let's not forget we are talking here about a SoC that
> nobody has actually designed yet.

 ah no... xilinx are designing one, it's called the zync-7000 series!

 l.