Discussion Intel current and future Lakes & Rapids thread

[Dayman1225]

That pretty much confirms the Digitimes rumor. You probably won't actually even see the dual core Cannonlake on shelves until Q3/4 next year.

You'll never see a CNL chip on shelves - they're Mobile chips 😉

 

[PeterScott]

Rumors?
It's the most recent event hosted by GloFo.
Shoo shoo shill.

I never said the event didn't happen.

Digitimes is an untrustworthy site. It has a long history of mis-reporting, mis-translating, creating all kinds of incorrect rumors they attribute to industry sources.

I am using Global Foundries themselves, from their own slides, that isn't subject to errors in reporting.

They clearly show 12LP Risk production is 1H.

 

[jpiniero]

You'll never see a CNL chip on shelves - they're Mobile chips 😉

Well, yeah.... they have to put them into laptops first. That's what I meant.

 

[Yotsugi]

Digitimes is an untrustworthy site.

This time they are reporting GloFo's event. It's not rumors, just quotes, questions and answers.

They clearly show 12LP Risk production is 1H.

The other slide from the very same deck said H1 volume so whatever.

 

[Dayman1225]

Well, yeah.... they have to put them into laptops first. That's what I meant.

Regardless - H2 is a 6 month period as you know. DigiTime said EOY 2018, which would imply Q4 Nov - Dec. We'll see when Intel actually puts 10nm into HVM, though this all depends on what Intel means by "low volume" of course, whether its non existent or that you can actually buy it. So we play the waiting game again.

 

[PeterScott]

This time they are reporting GloFo's event. It's not rumors, just quotes, questions and answers.

The other slide from the very same deck said H1 volume so whatever.

Show me that slide then.

 

[Yotsugi]

Show me that slide then.

Too lazy to do it (and you can do it yourself if you try really hard).
Anyway for company that was touting their node advantage for ages letting the schedule slip so hard is unforgivable and you know it.

 

[PeterScott]

Regardless - H2 is a 6 month period as you know. DigiTime said EOY 2018, which would imply Q4 Nov - Dec. We'll see when Intel actually puts 10nm into HVM, though this all depends on what Intel means by "low volume" of course, whether its non existent or that you can actually buy it. So we play the waiting game again.

If CNL is only dual core parts it hardly matters. The market is really moving on to four cores. It makes CNL look like an extended 10nm test run.

 

[Yotsugi]

It makes CNL look like an extended 10nm test run.

Because it is one. Always was since they cancelled server Cannonlake.
10 performs worse than 14++, only 10+ catches up on performance.

 

[PeterScott]

Too lazy to do it (and you can do it yourself if you try really hard).
Anyway for company that was touting their node advantage for ages letting the schedule slip so hard is unforgivable and you know it.

I have looked and could find nothing, I only found the slide I linked from multiple sources. More likely you are just "misremembering".

 

[Yotsugi]

I have looked and could find nothing, I only found the slide I linked from multiple sources.

Try harder, dude. You can do it (hint: it the slide with both 7LP and 12LP)!

More likely you are just "misremembering".

Oh no, I am very, very bad at forgetting things.

 

[krumme]

Try harder, dude.
.....

Oh no, I am very, very bad at forgetting things.

Are you a woman?

 

[Yotsugi]

Are you a woman?

I am a shitposter, and a pretty decent one at that.
But 10 going HVM H2 2018 is bad no matter how locals might try to spin it out.

 

[krumme]

I am a shitposter, and a pretty decent one at that.
But 10 going HVM H2 2018 is bad no matter how locals might try to spin it out.

I try a spin too then.
Its the same dudes making the node as yesterday. Why pour more money into speeding up 10nm progress?

I simply dont see the business advantage.
I know Intel management have been reciting the moores law like autist with intellectual functional deficit but adapting capex seems more sane than pouring more x86 wonder tablets on the market to beat apple samsung google and the rest of the arm ecosystem.

 

[Yotsugi]

Why pour more money into speeding up 10nm progress?

Because Intel tried to play foundry business since around 2014(?). They've lost nearly all design wins for 10nm because of delays.

What's worse, TSMCs 7nm for SoCs is going HVM H1 2018, and GloFo is preparing an actual high performance 7nm node for H2 2018.

They've lost the war for bleeding edge nodes and now FD-SOI is about to hit from the other direction.

 

[krumme]

Because Intel tried to play foundry business since around 2014(?). They've lost nearly all design wins for 10nm because of delays.

What's worse, TSMCs 7nm for SoCs is going HVM H1 2018, and GloFo is preparing an actual high performance 7nm node for H2 2018.

They've lost the war for bleeding edge nodes and now FD-SOI is about to hit from the other direction.

Yes and thats why i think their response is right. They cant beat iphone money and ibm basic research so getting for right process node and cost is better than braindead leading stuff without a purpose.
I am sure there is some people at Intel that still think they are leading the process stuff and can be a foundry business. Thats what monopolies are made of.
The dangerous stuff here is if we have a new capex alert. That future margins can not uphold all the really fixed cost. Thats stuff thats can keep you awake at night.

That and writing on forums ofcource...

 

[Yotsugi]

so getting for right process node and cost is better than braindead leading stuff without a purpose.

You want cost-optimized node? That's Samsung or TSMC territory. You want low leakage or RF stuff? That's GloFo's territory.
Intel's bread and butter were always bleeding edge nodes and they've lost their advantage.

I am sure there is some people at Intel that still think they are leading the process stuff and can be a foundry business.

These people are somewhere in the upper management.

 

[krumme]

You want cost-optimized node? That's Samsung or TSMC territory. You want low leakage or RF stuff? That's GloFo's territory.
Intel's bread and butter were always bleeding edge nodes and they've lost their advantage.

These people are somewhere in the upper management.

Its only bread and butter so long you have the portfolio to fund it. 4 years ago that was easy to see. Today. Not so much.

The 10nm server was actually such a product. And granted thats where it hurts.

There are bound to be development errors. Euv adaption to late what not. But its easy to pinpoint in hindsight.
I dont know. But if you buy 10 new big lithography machines and use several B to play with it just to learn you better have a good and some sort of specific business case imo.

And whats that case today?

 

[raghu78]

You want cost-optimized node? That's Samsung or TSMC territory. You want low leakage or RF stuff? That's GloFo's territory.
Intel's bread and butter were always bleeding edge nodes and they've lost their advantage.

These people are somewhere in the upper management.

Very well said. TSMC 7nm will beat Intel 10nm in time to market with much higher volume (Apple 2018 iPhone) in 2018. Intel's process speciality was high performance. They sucked at low power and high density against TSMC or Samsung. Intel processes never ever translated their superiority from technical specifications to actual products. Its no wonder why Intel never releases transistor count and die size info for Atom as that would draw comparisons against low power ARM SoCs from Apple and Qualcomm

Its only bread and butter so long you have the portfolio to fund it. 4 years ago that was easy to see. Today. Not so much.

The 10nm server was actually such a product. And granted thats where it hurts.

There are bound to be development errors. Euv adaption to late what not. But its easy to pinpoint in hindsight.
I dont know. But if you buy 10 new big lithography machines and use several B to play with it just to learn you better have a good and some sort of specific business case imo.

And whats that case today?

Are you kidding. Intel's 100% x86 server market share is dependent on them behind ahead of AMD with superior products (based on architecture and process node). Intel's 10nm delays are now reaching disastrous levels. 10nm was expected in late 2016 and now we are not even sure if it will arrive in late 2018 with real high volume. If Intel cannot ramp CNL in volume in H1 2018 what are the chances that 10+ which builds on the yield learning at 10 will ramp in H1 2019. Intel's entire PC and server business is dependent on them behind ahead of AMD. The situation can get very ugly very quickly for Intel. AMD is ramping multiple products across desktop, notebook and server and its executing quite well. GF also is showing signs of improved execution with 12LP products like Pinnacle Ridge arriving in Q1 2018. GF 7LP is scheduled for risk production in H1 2018 with a ramp to HVM in H1 2019. If Intel does not fix its 10nm process problems within the next 12 months and ramp 10+ in H1 2019 they are in serious risk of falling behind AMD and we could see the unthinkable happen - Zen 2 arriving before Icelake. If that were to happen we will see a tipping point where Intel's business will start to crash rapidly starting from 2019 and Intel will be forced to shut down fabs.

 

[Yotsugi]

HVM in H1 2019

7LP is going HVM H2 2018, which makes the situation a little bit more fun...
...for AMD, that is.

and we could see the unthinkable happen - Zen 2 arriving before Icelake

Brb gotta prepare my salty tears bucket.

 

[TheGiant]

and we could see the unthinkable happen

Imagine that. Intel end 2018, no new mobile, server or desktop core and the same 14nm ++++++++ node...

WTF is happening there? Are all the engineers gone and just the "gief more meetings and sweet charts of self defense managers" left?

 

[Yotsugi]

Are all the engineers gone and just the "gief more meetings and sweet charts of self defense managers" left?

Yes.
Intel died with Otellini.

 

[CHADBOGA]

Imagine that. Intel end 2018, no new mobile, server or desktop core and the same 14nm ++++++++ node...

WTF is happening there? Are all the engineers gone and just the "gief more meetings and sweet charts of self defense managers" left?

It's all about "The Diversity" now

 

[DrMrLordX]

If only they had listened to Andy. If only they were more paranoid.

 

[Yotsugi]

If only they had listened to Andy. If only they were more paranoid.

Now they can listen to ghost of Otellini and bribe the shit out of every OEM on the market.