Maynard Handley (name99.delete@this.name99.org) on October 2, 2013 5:59 pm wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on October 2, 2013 4:17 pm wrote:
> > > The results don't lie. Intel has a big lead in process and
> > > has for a long time. You might handwave and say that
> > > the gap is going to close, but so were people saying the same thing 1 year ago, 2 years ago, 3 years ago.
> > >
> > > So if there is evidence that the gap actually shrinks, then it will be something
> > > to take notice of. Roadmaps and announcements aren't worth much.
> >
> > I've been watching the industry carefully since 2002. For over a decade I've heard pronouncements
> > that 'other people' will catch up with Intel...and for over a decade it's been rubbish.
> > Initially, it was AMD/Motorola/IBM, and now it's TSMC or Samsung.
> >
> > Either way, I don't believe the gap will narrow...although I'm willing to re-evaluate
> > that view every few years. I sure don't see it for 20nm or 16nm.
> >
> > David
>
> I don't view this as a very enlightening quarrel, but I will point out that, even worse than
> unenlightening, it is meaningless in the absence of a definition of "catch up with Intel".
> There are MANY (perfectly reasonable) definitions of "catch up"
> and for many of them, people HAVE "caught up" with Intel.
> - Yes, if the metric is "sell a faster x86 than Intel"
> sure, no-one's going to catch up. BUT if the metric is
> - sell a faster (or higher performance/watt) CPU in the mobile space ... (Apple)
> - sell semiconductors of higher total value ... (TSMC)
> - sell a more monstrous server CPU ... (IBM)
>
> Alternatively your metric might be the very specific "minimum feature size". OK, Intel wins that. But
> an alternative metric is transistor density and certainly Samsung, probably TSMC, beat Intel.
>
> Intel has great process, sure. They're generally first to use exotic materials and exotic structures. That's
> great, and you're welcome to say that you expect them to continue being first in those areas. But that is not
> the same thing as "winning". This is kinda the whole point of this thread --- that there's a whole universe
> of dimensions along which a company in this space can excel, and Intel does NOT dominate all of them.
>

Intel will "win" if it is very good at designing high performance per watt processors, has a much better than everyone else fab, and will make the right business decisions.

I doubt anyone will argue that Intel is unquestionably the best at numbers 1 and 3 in that list. They have a lot going for them in number 1 and I doubt they're stupid, but the same is true for a lot of their competitors. But number 2 is really not in question right now and will not be for at least another 3 years.

The argument in question for this part of the thread was whether Intel has a "better" process, by any reasonable definition of the word better; not who makes the better chips, or makes more money. And no one has had a better process than Intel for quite a long time. I don't think that Haswell and the A7 are at all comparable in terms of what we can expect for transistor density. Secondly, we don't even know whether transistors were referring to physical transistors or layout transistors in these cases. Realistically, I don't believe that a process with tighter contacted gate pitch, tighter minimum metal pitch, and smaller minimum SRAM cell size has a *lower* equivalent transistor density than Samsung's process.