[Extech]Intel's 14nm milkshake

[jpiniero]

@jpiniero - if none is buying it, why would they introduce retina display, bring in crystall well, tout about improved battery life. up until two years back, they held the highest market share in laptops sold above $1000. why would they jeopardize that

I didn't say Apple will get out of the laptop business - just release replacement products running iOS once OSX gets merged into it. Haven't seen anything about what res the 12" iPad will be at, but it could be 4k.

 

[hawtdawg]

Maybe Intel sees the writing on the wall, and decided that the best thing to do is push forward as the worlds premiere fab. Im sure there is all sorts of money to be made if every SoC developer on the planet wants to use your fab process.

 

[Homeles]

but shouldn't 14nm be smaller than 16nm by ~12.5% (2/16), so the gap is only 20% or so it would be according to their claims. I am not sure if a few nm will have huge effects on power but i guess we will see.

It seems you haven't been participating in recent discussions on this forum. That's okay, I'll bring you up to speed.

14nm, 16nm, 28nm, etc. are marketing labels, and are decoupled from feature size. They used to indicated minimum feature size, but that is no longer the case anymore.

my only issue with the analyst day was intel opening its fabs up for other competitors. Id rather not have them enable nvdia/qcom. Apple i guess is fine since they dont really compete head to head. But giving qcom access to INTEL's best assets seems awful awful awful strategy

They aren't going to open it up to Nvidia and Qualcomm.

Apple abandoned the Pros a long time ago. Even the new Mac Pro is really a halo device - they don't expect anyone to actually buy it.

I'm not entirely sure what planet you're from, but here on Earth, companies generally sell products that they expect people to buy.

 

[ShintaiDK]

I didn't say Apple will get out of the laptop business - just release replacement products running iOS once OSX gets merged into it. Haven't seen anything about what res the 12" iPad will be at, but it could be 4k.

Apple said, not long ago, that IOS and OSX would not merge.

The disaster at Microsoft is actually due to trying that. And I am sure everyone in the top would point that out, anytime someone comes up with the idea. Ballmer and Sinofski got fired due to that. Number 1 and 2 in MS.

 

[jpiniero]

Apple said, not long ago, that IOS and OSX would not merge.

You only have to look at the changes that Apple has made in the newer versions of OSX to see where things are going. They are trying to be more deliberate about it than, say MS.

 

[6PPC]

Maybe Intel sees the writing on the wall, and decided that the best thing to do is push forward as the worlds premiere fab. Im sure there is all sorts of money to be made if every SoC developer on the planet wants to use your fab process.

That begs the question can they produce a SoC's for phones. The new modem chip announced by Intel, SoFIA, is manf. by TSMC and will be for the next 2 years. It would seem that Intel is unable to manf. the right silicon for phones.

 

[24601]

What are the chances of Nvidia and AMD/ATi fabbing at Intel 14nm? (Or 10nm)

 

[meloz]

I find the yield data most interesting. Even though one of the axis is not graded you can tell 22nm started very strong, 14nm by comparision seems like an utter disaster. But they have managed to reclaim the lost yield very quickly, with major breakthroughs in July-August and steady incremental improvements ever since. Utterly Amazing.

So the makes the decision to launch Haswell-Refresh looks even more perplexing. The yield is there to launch Broadwell for both laptops and desktop at the same time. Likely the decision to launch refresh has more to do with keeping 22nm fabs occupied than anything, and thus obtaining the highest possible profit margin.

 

[ShintaiDK]

What are the chances of Nvidia and AMD/ATi fabbing at Intel 14nm? (Or 10nm)

About 0. And for the same reason why they are last in the 20nm race. And CPU wise for AMD also why Kaveri comes on 28nm in feb.

 

[ShintaiDK]

I find the yield data most interesting. Even though one of the axis is not graded you can tell 22nm started very strong, 14nm by comparision seems like an utter disaster. But they have managed to reclaim the lost yield very quickly, with major breakthroughs in July-August and steady incremental improvements ever since. Utterly Amazing.

So the makes the decision to launch Haswell-Refresh looks even more perplexing. The yield is there to launch Broadwell for both laptops and desktop at the same time. Likely the decision to launch refresh has more to do with keeping 22nm fabs occupied than anything, and thus obtaining the highest possible profit margin.

You assume the graphs spans a different timeframe for each one? It shows the same timeframe for both.

There is nothing on that graph that tells you why there is a Haswell Refresh on the desktop. Airmont is the reason for it, not yield or the node progress.

It simply changed from server/mobile/desktop at edge node, atom at secondary. To server/mobile/atom at edge node and desktop secondary. in other words, limited capacity.

 

[Yuriman]

You assume the graphs spans a different timeframe for each one? It shows the same timeframe for both.

There is nothing on that graph that tells you why there is a Haswell Refresh on the desktop. Airmont is the reason for it, not yield or the node progress.

It simply changed from server/mobile/desktop at edge node, atom at secondary. To server/mobile/atom at edge node and desktop secondary. in other words, limited capacity.

I believe he means the Y-axis. If you notice, there's no scale and it doesn't start at zero.

 

[ShintaiDK]

I believe he means the Y-axis. If you notice, there's no scale and it doesn't start at zero.

Nope.

Even though one of the axis is not graded you can tell 22nm started very strong, 14nm by comparision seems like an utter disaster

There os no data on the graph you can use to compare the development of the 2. Its simply here and now graphs.

They estimate (2 year old) 22nm and 14nm will same the same yield in Q1 2014.

 

[SiliconWars]

They estimate (2 year old) 22nm and 14nm will same the same yield in Q1 2014.

And they won't move the process until that point. Even if yield matches in Q1, going on the history of 14nm they'll want to ensure there are no sudden dips again, so that's another month or two to make sure the process is perfect.

So we're at ~6 months late as I said, and quite possibly longer. Ain't no way to get a quarter delay only out of that information.

 

[Khato]

That begs the question can they produce a SoC's for phones. The new modem chip announced by Intel, SoFIA, is manf. by TSMC and will be for the next 2 years. It would seem that Intel is unable to manf. the right silicon for phones.

Intel's Infineon acquisition was only completed at the beginning of 2011, so not quite three years now. Since then I believe it's safe to assume that they've been working on their LTE modem pretty much non-stop. Now sure, Intel could have told them to change the process target to their internal fabs... but how much would that have delayed introduction? Best case would be two quarters, and more likely it'd be up to a year just because of the different methodologies involved. No, it makes far more sense to get the product out first and then transition production to their own fabs.

I find the yield data most interesting. Even though one of the axis is not graded you can tell 22nm started very strong, 14nm by comparision seems like an utter disaster. But they have managed to reclaim the lost yield very quickly, with major breakthroughs in July-August and steady incremental improvements ever since. Utterly Amazing.

Have to remember that the yield data chart doesn't go back very far at all - it's basically only two quarters worth of data. Whereas according to the rumors Broadwell taped out what, well over a year ago? So it's likely that if you go back further on 22nm it'd look just as bad as Broadwell, it's just that Broadwell didn't really start improving until recently.

Other thing to keep in mind is that the ungraded Y axis very well be on a logarithmic scale.

 

[jhu]

 

[R0H1T]

The question might be, who is willing to pay more than they even do at TSMC. Apple is certainly one of them. And Apple, while competition, is not in a direct competition.

Apple likes their margins as much as, if not more, than Intel itself & so neither Apple nor Qualcomm is going to their biggest competitor's fabs anytime soon. Actually Intel opening up their fabs proves one of two things ~

a) They realize how hard it's getting to keep their tick-tock cycle in place & also seeing how they've had yield issues with their 14nm Intel now wants to fund their node shrinks with a little help from their competition in the mobile space. The biggest obstacle for them however is that the top three in the mobile/tablet arena really do not need Intel at all & this is because ~

1) Apple is shifting towards TSMC gradually, away from SAMSUNG for obvious reasons, so I doubt Intel would ever make the Apple custom SoC as cheap as TSMC or even SAMSUNG & then there's rumors flying around that GF could supplement SAMSUNG in their chip making endeavor for Apple.

2) Qualcomm fabs their chips at TSMC & SAMSUNG, correct me if I'm wrong here, so I doubt they'll need Intel anytime soon seeing how both of these foundries are in top gear for their next round of node shrinks, plus Intel being their biggest potential competitor in the mobile/tablet arena would make it that much hard for'em to justify the shift.

3) SAMSUNG doesn't need Intel at this point in time & potentially could rival them in the fab biz as they have the knowledge & money to pull this off, the only mega conglomerate aside from IBM that can do so in the not so distant future.

b) The revenues from their mobile division are barely trickling in atm & it could be that their internal projections are somewhat pessimistic so they'd like to open up other potential avenues for growing their top & bottom lines. Again I don't see how any of the major chipmakers will go to Intel in the near future unless they're pretty desperate for volumes & are willing to take a hit to their margins in the process!

The reason I've highlighted these points is because a while ago, on a similar thread, nearly everyone agreed that Intel would "never enable their competitors come what may" so that just got me thinking that either Intel has assessed the tough times ahead are worse than they originally thought or they're just trying to make a quick buck. Either way, depending on the response they get, Intel is still kinda desperate(seeing how their core PC biz is tanking) & seemingly they wanna diversify, only this time with conviction.

 

[jdubs03]

If there is any company that will decide to go to the premier node it would be Apple, though it wouldn't be an overnight process, aka 2015-2016 time-frame so 10nm.

I don't think Samsung will catch up; and even if they did, if Apple is relying less and less on Samsung the more they will have to rely on GloFo and TSMC, two foundries that are even more behind than Samsung, plus as seen in the Intel slide, the density advantage for them is now a disadvantage.

Apple would have to pay a higher price, but that is to be expected, we're talking about potentially skipping an entire process generation. Would it be worth it?

All in time...

 

[Homeles]

You assume the graphs spans a different timeframe for each one? It shows the same timeframe for both.

"Yield at the same point in development."

 

[mikk]

I believe he means the Y-axis. If you notice, there's no scale and it doesn't start at zero.

Both started zero though.

And they won't move the process until that point. Even if yield matches in Q1, going on the history of 14nm they'll want to ensure there are no sudden dips again, so that's another month or two to make sure the process is perfect.

So we're at ~6 months late as I said, and quite possibly longer. Ain't no way to get a quarter delay only out of that information.

Yield parity to 22nm is predicted for early Q1 2014 with mass production start following later in Q1 targeting a Q3 launch. We are not 6 months late, instead we are 3 months late.

 

[Arachnotronic]

Both started zero though.





Yield parity to 22nm is predicted for early Q1 2014 with mass production start following later in Q1 targeting a Q3 launch. We are not 6 months late, instead we are 3 months late.

[redacted]

Personal attacks will not be tolerated
-ViRGE

 

[SiliconWars]

woops

 

[SiliconWars]

Yield parity to 22nm is predicted for early Q1 2014 with mass production start following later in Q1 targeting a Q3 launch. We are not 6 months late, instead we are 3 months late.

So last minute Q1 with last minute Q3 launch? k.

 

[jpiniero]

We are not 6 months late, instead we are 3 months late.

I'll say October for the QC HQ BGA models, November for Airmont, December for the Broadwell-K and Xeons and January and February for duals and the rest of the quads.

 

[SiliconWars]

I'll say October for the QC HQ BGA models, November for Airmont, December for the Broadwell-K and Xeons and January and February for duals and the rest of the quads.

Smart money says Intel doesn't care about anything that isn't Atom, still. The ARM guys are still a node or two away in competing with x86 in servers. Broadwell has no competition and can be left to lag.

Intel needs to put all their effort into Atom, 2 years ago.

Process just isn't that important any more. What Intel can offer vs the rest isn't worth the cost. Really simple stuff.

 

[mikk]

So last minute Q1 with last minute Q3 launch? k.

I don't know exactly when in Q1 production starts but even a late Q1 production start makes an early Q3 launch possible. As a general rule 3 months or 4 from production to launch is a usual timeframe. Of course I fully understand that you are hoping Intel is later than this.