Richard Cownie (tich.delete@this.pobox.com) on February 1, 2013 10:11 am wrote:
> anon (anon.delete@this.anon.com) on February 1, 2013 9:36 am wrote:
>
> > Well I believe it is you who do not understand. Their dominance is in manufacturing.
> > They achieved that dominance in logic mfg with their x86 market, certainly. From there,
> > they have been able to plunder much of the RISC and Unix server markets, which were
> > absolutely not x86 dominated, and are actually notoriously adverse to change.
>
> Well, not *that* adverse to change.

But apparently you think the Android market is :)

In fact RISC/UNIX servers are *far* more resistant than the Android market.

> The "RISC market" that Intel plundered from
> 1995 on didn't even exist in 1985. Change doesn't happen overnight, but it
> certainly does happen over 5-10 years. And the increasing use of open-source
> software has made it much easier to switch to a different ISA.
>
> And for the relative importance of manufacturing vs design, you could look at
> iAPX432 or i860 or Itanium, all products which had access to Intel's
> manufacturing but either vanished without trace, or limped along without
> ever establishing dominance.

None of them had the correct volume/profit ratios, it's not a real mystery. It's not a magical "intel manufacturing advantage steamrolls everything" effect. It is "intel has been the best at competing in, and has the best manufacturing technology to suit large volumes". Besides, they were also significantly competing against Intel x86.

>
> > > Outside the desktop/laptop business - which is now shrinking - points a/b/c don't
> > > apply. I don't think manufacturing excellence alone will allow Intel to dominate
> > > other markets. You have a different opinion, evidently. We'll see.
> >
> > Neither did they apply in the server market.
>
> Well, the "server market" is not uniform. The network effects are still going
> to apply quite strongly to things like enterprise database software. But there's
> a heck of a lot of "servers" running a Linux/Apache/MySQL/PHP stack, and none of
> that is particular tied to x86 in the way that WindowsNT was (and my recollection
> is that Linux was also mostly for x86 back in 1995-2000 - and also didn't
> match the quality of proprietary Unix'es back then - though obviously these
> days it's fine on ARM and just about anything else).

These all took over from entrenched RISC/UNIX, though.

>
> > The key is that the volume of silicon, and the revenue, needs to be large enough
> > that it suits Intel's manufacturing model. That is: high spending on fixed costs
> > in order to reduce the variable costs. Server market has this property.
>
> Not really.

Yes really.

> 10M server cpu's a year is nice - and maybe those are 250mm2 each,
> compared to 50mm2 for a phone/tablet chip. But the phone/tablet chips sell
> in the hundreds of millions - so that's a bigger driver of manufacturing volume.

I keep repeating to you: revenues. Intel doesn't give a crap about microcontroller market that ships billions of devices.

>
> Intel does well so far because they've got about 85M units of desktop/laptop to
> drive their volume. But that is a) not as much silicon as phone/tablet chips
> and b) shrinking. The shrinking of Intel's core market, and the growth of the
> phone/tablet market - in which Intel is not (yet) a significant player, are the
> important trends.

Yes, I keep repeating myself. If the market changes and smartphone becomes significantly high volume and high revenue to matter for Intel, then they will be competitive there and entry to ARM market really is not as big a problem as you're trying to make out.