[Arm-netbook] Rpi vs cortexA8

[lkcl luke]

On Sun, Jun 3, 2012 at 11:42 AM, Enrico <ebutera at users.sourceforge.net> wrote:

> Just speculating, but one of the things that could give a speed boost
> in the cortex-a8 is the neon unit.

 my associate geoff worked for LSI Logic.  ARM approached LSI Logic a
good couple of decades ago, saying that they'd designed a super CPU,
very low power, and could they please take a look?  they were told, in
a nutshell, "it's shit".  the design was so full of bottlenecks the
LSI Logic team didn't even know where to begin to describe the
problems.  ARM didn't believe them.

anyway, it turned out that LSI Logic's team were spot on: as the CPU
went through more and more revisions, they had to put more and more
patches in place to work around the bottlenecks, but the CPU's design
is so fundamentally flawed that there are hard limits on how fast the
architecture could go.

the Cortex A8 however was *not* designed by ARM.  it was designed
based around a streamlined pipeline architecture (Harvard
Architecture) right from the word go, by a VC-funded team in Austin,
Texas, that ran out of money.  when they ran out of money they
approached ARM and asked them if they'd like to buy the company, which
they did, because ARM's own design was so shit.

the Intel PXA designs (bought by marvell) are a different story.  ARM
was desperately short of cash (almost always) so they did a deal with
Intel where Intel would be granted access to the ARM design, would pay
them a fixed one-off royalty for doing so (!) but would *also*
contribute *back* any design changes to ARM.

Intel's team clearly took one look at the design and fell off their
chairs laughing, because what they actually did was a total redesign,
similar to what that team in Austin Texas did, some years later, using
again a pipelined architecture behind-the-scenes and adding in
superscalar (out-of-order execution) etc. etc.

the redesign was so radical - out of necessity - that Intel *refused*
to contribute it back to ARM, because it was just an "interoperable"
design that was, in essence, a complete and total redesign and so
nothing to do with ARM!

anyway the point is that there are at least three totally different
architectures, here, one of which (designed by ARM) is extremely poor,
and two of which are actually very good.  comparing them is like
comparing... the Motorola 68040 to a.. a... i dunno, pick another
obscure CPU, they're that different.  the only thing in common is that
they can execute the same instructions (almost).

l.