dmcq (dmcq.delete@this.fano.co.uk) on August 20, 2014 8:17 am wrote:
> Jouni Osmala (josmala.delete@this.cc.hut.fi) on August 20, 2014 4:33 am wrote:
> > > > I agree on you with ISA effect between ARM and Intel, but for process there is
> > > > no-one like Intel, and that's the strong enough force that makes cows fly.
> > > > They put more money on process development than anyone for
> > > > decades they probably got the best process engineers
> > > > since everyone in the field knows if you want to be ahead of everyone else go to Intel. There are plenty of
> > > > parameters to tune with different trade offs then you should
> > > > consider the pipeline between starting to develop
> > > > a manufacturing process to having product at the hands of
> > > > the customers, that pipeline is quite long and people
> > > > make announcements at different stages of that pipeline
> > > > and you have clearly mistaken those stages. Intel got
> > > > improved transistors in their 22nm process and now others are
> > > > aiming in their 14nm process to have same improvement
> > > > in their transistors, but Intel got nice improvement over
> > > > their 22nm transistors. So in the end, the manufacturing
> > > > process might actually be the thing that keeps x86 ahead of ARM indefinitely.
> > > >
> > > > As for delay of Broadwell what I think has happened is that 22nm has really good yields and
> > > > 14nm yield improvement curve changed temporarily its trajectory to not improving as fast as
> > > > traditionally it has and they had to put on hold the retooling for 14nm process, and then they
> > > > got it back few months later. And that was quite close to point in which they would start converting
> > > > for mass production but just below it. Now if everyone else has same problem, then there is
> > > > bigger delay between some production and high volume complex chip than normally.
> > > >
> > >
> > > Yes, Intel has shined in manufacturing and it still has a leading gap over rest of foundries, but my point
> > > is that the gap has been reduced in recent times. E.g. Broadwell-EP on 14nm FinFET is scheduled for Q4 2015
> > > and ARM SoCs on 16nm FinFET are scheduled for 2016. Intel will have only a half-node advantage this time.
> >
> > Those ARM SoC:s probably are more Comparable to Broadwell-Y than Broadwell-EP in
> > terms of how good defect density you need in order to get reasonable yield.
> >
> > What different players have disclosed from their 14nm&16nm processes Intel is having significantly
> > better 14nm process than anyone else. Others probably need their 10nm process to compete
> > with Intel 14nm process. Actually the 14nm intel process is such advantage over competition
> > that it just made me scared that future might become all Intel.
> >
> > Everyone else is investing heavily to get the features Intel already has in its process
> > while Intel is investing heavily to get improved version of it, that is Intel is reaping
> > the benefit of momentum it has already build up in its process development. Others are spending
> > billions to get features Intel spent billions on previous generation process.
> > Also TSMC needs to spend more to get all the variants of its process done and to
> > get higher momentum than Intel to catch up but we don't see that happening.
>
> What I see is that a few years ago Intel introduced the crappy Atom with a load of fanfare saying
> it was going to crush ARM, and it looked like a very reasonable competitor - but it didn't compete.
> Last year it introduced the far better Bay Trail processors to crush ARM but is spending more in
> a quarter on 'contra-revenue' and getting even less revenue. And next year it looks like it is putting
> variants of its Haswell architecture into the fray. Will Intel have to give contra-revenue away with
> its Haswell class cores? And in 2016 or 2017 will they be giving contra-revenue in servers as well
> and what will that do to their contra-contra-revenue and prospects for the future?

You totally missed the thing I was saying.
ARM architecture is CHICKEN, X86 architecture is PIG. Manufacturing process is FORCE.
You see both fly, which one has bigger force behind it?
I was arguing that no-one can compete with Intel on how much force they can put behind their chips, and that force has increased in alarming way. Atom is an experiment on how low force you can put behind PIG and still get it fly. As with servers they are getting much more force behind it.

At atom power level the decoders are quite critical limitation, but with bigger density and power consumption improvements in their manufacturing process than with other foundries, they are gearing towards high end x86 cores in smartphone power consumption levels some years. That's what I consider alarming because I don't want to get stuck with a pig for rest of my life.