[Arm-netbook] Slowly but surely...

[Gordan Bobic]

On 07/09/2012 02:41 PM, lkcl luke wrote:

[openmoko being wrong example]

> in this case, all that happens is we move to a different CPU.

In which case you still need a new PCB in the module. Do we have a 
working PCB for the A10 yet? How long did that take so far? I've been on 
this list for about a year now. That's a long time.

> that's always been the strategy.  remember: i'm talking to several SoC
> companies in order to plan the roadmap ahead for the next 4-5 years,
> and the possibility has just come up to work even beyond that
> timescale.

Which would be a lot more meaningful if there was actually a tangible 
product to plan this future for. At this point, given the rate of 
progress so far, I am not even 100% convinced there will even be an EOMA 
card with an A10 in it.

>   we know that the first CPU Card is going to be the toughest.  getting
> parts.  finding suppliers.  getting I/O Boards made.  convincing
> clients to use it.  getting the software in place.  modifying the
> linux kernel to support EOMA.  it all has to happen at once.

And it is increasingly looking like the ship that the A10 EOMA needed to 
be on has sailed. The optimist in me hopes that proverbial ship may 
still be in swimming distance, but the realist in me thinks there's a 
snowflake's chance in hell of making it. As you say - it's a lot of work 
to do.

>   once even _one_ board, _one_ CPU card, _one_ product is done, any
> others are easier.  why? because the design is modular. why does that
> matter? because you do *not* need a total redesign of *any* products,
> just to put in a new CPU Card.  why? because the software comes on the
> CPU Card.

Sure. To chance the SoC you "only" have to change the EOMA card PCB and 
the kernel. That is again a fair amount of work.

These new blade servers have the same advantage in the server context as 
the CPU cards are pluggable.

The main difference is that if you ask very nicely and get in with the 
right crowd, you can plausibly get your hands on one of those server 
chassies - unlike with EOMA.

>   so even if the EOMA-68 A10 CPU card is late as hell, it *doesn't
> matter*.  why? because it will be a critical prototype component
> against which all the *other* pieces of the strategy can be made
> concrete.

Except if you chance something as critical as the SoC, you'll have to 
redesign the PCB anyway to make all the lines go to the right place.

>   so do you see, alexey, how completely different this is from the
> *one* openmoko product?  do you see how comparing *one* specific
> (failed) product to an entire product design *strategy*, of which the
> A10 EOMA-68 CPU Card is just one small part, is not comparing
> like-with-like?

Can't speak for Alexey, but to me it looks like you are arguing the 
wrong point. You are arguing that you can reuse the machine running 
based on the EOMA, not the EOMA module design itself. Sure, you are 
making it a little easier for the end user, but it doesn't mean you 
don't have to redesign the EOMA PCB and get the new kernel working every 
time you chance the SoC. In reality, in the server context, the same 
applies to these new lines of ARM servers, so EOMA doesn't really have 
any obvious advantage in that context.

Gordan