Discussion Intel current and future Lakes & Rapids thread

[TheGiant]

is this all because Brian Krzanich happened?

 

[liahos1]

10nm icelake was delayed from late q2 to early q3. It was always scheduled to come out after cooper lake. I don't see how any of this is new news.

 

[Yotsugi]

They could just cancel it and release Sapphire Rapids, which you would think would be easier to yield

They're barely yielding mobile stuff, which contains easily binnable iGP.
SPR won't be any easier.

 

[jpiniero]

SPR won't be any easier.

It's chiplets connected via EMIB, is it not? It should be easier.

 

[liahos1]

Yotsugi where are you getting your sourcess? Charlie?

 

[DrMrLordX]

I bet they ship binned HCC by late'20 so that shareholders won't crucify them.

That is always a possibility. Intel has already set the bar low by promising 36c dice or whatever it was the last time they said anything about it.

They could just cancel it and release Sapphire Rapids, which you would think would be easier to yield.

If they build Sapphire Rapids out of EMIBed 4c dice then yes, I think it would be easier to yield. It should be interesting to see how that works out.

is this all because Brian Krzanich happened?

He probably isn't the only one responsible. Their foundry is in shambles. The 14nm delays followed by 10nm failures led to the current state of Intel's progress.

10nm icelake was delayed from late q2 to early q3. It was always scheduled to come out after cooper lake. I don't see how any of this is new news.

IceLake-SP was supposed to be Q4 2019 or Q1 2020. It won't be Q1 2020 anymore.

 

[Yotsugi]

It's chiplets connected via EMIB, is it not?

Don't think so.
Granite should be Foveros + EMIB, though.

Yotsugi where are you getting your sourcess

These are my personal ones.

has already set the bar low by promising 36c dice or whatever it was the last time they said anything about it.

26c HCC, 38c XCC.

 

[jpiniero]

Don't think so.

That was the rumor I had seen; Sapphire uses EMIB while Granite uses Foveros.

 

[Yotsugi]

That was the rumor I had seen; Sapphire uses EMIB while Granite uses Foveros.

Granite uses both Foveros and EMIB, that's the big bet.
Dunno how it'll pan out, but good luck to our friends at Intel.
Also I'm pretty sure SPR is also PR quantities only.

 

[Cardyak]

It appears Intel's 10nm woes are near infinite and it will never be a real viable product. This is doubly irritating, not just due to lack of competition, but more because their microarchitecture is top notch but unavailable on desktops.

Their only hope as far as I can see is to bet the farm on 7nm and get Ocean Cove out ASAP.

It really is last chance saloon now for Intel as far as process is concerned, they need 7nm to clock highly, have high yields, and be economically viable to justify the existence of their fabs.

A lot rests on Jim Keller's shoulders now, Intel's next generation architecture needs to be the Core architecture of the 2020's - Something that can offer a massive leap in IPC and also act as a foundation for which to build the next decade of innovation. In the end I expect this pressure will actually amount to increased innovation, if Ocean Cove is indeed Intel finally wiping the slate clean and creating a brand new x86 implementation from scratch then I expect it to be an absolute monster in terms of IPC.

 

[Yotsugi]

A lot rests on Jim Keller's shoulders now, Intel's next generation architecture needs to be the Core architecture of the 2020's

Jim can't magically fix their execution or fractures in their corporate culture.
Intel now has to do what AMD did: bet big and hope it turns out even better!

 

[liahos1]

lol this thread is amazing. some random source says desktop product delayed till 2022 on 7nm and the intel is doomed patrol comes out swinging.

Again, my sources say wafer starts are set to double this month at 10nm. why do that if the process is loss making, broken irrevocably etc

 

[Yotsugi]

lol this thread is amazing. some random source says desktop product delayed till 2022 on 7nm and the intel is doomed patrol comes out swinging.

Again, my sources say wafer starts are set to double this month at 10nm. why do that if the process is loss making, broken irrevocably etc

Is this some kind of extravagant coping mechanism?
It's cute I give you that.

 

[Thunder 57]

He probably isn't the only one responsible. Their foundry is in shambles. The 14nm delays followed by 10nm failures led to the current state of Intel's progress.

Maybe they hired some GloFo people on the cheap to run their fabs awhile back?

Jim can't magically fix their execution or fractures in their corporate cultures.
Intel now has to do what AMD did: bet big and hope it turns out even better!

I don't get why people think Jim Keller is some wizard that can fix any problem anywhere. I'm sure he is a super talented guy, but no one person is going to fix a problem this large.

 

[liahos1]

Yotsugi, just because you and Charlie says something is so, doesn’t make it so.

That isn’t a coping mechanism. That is basic logic. You have your “sources”. What do you do for a living?

 

[jpiniero]

Yotsugi, just because you and Charlie says something is so, doesn’t make it so.

Pretty obvious they "delayed" (read: it's getting cancelled) Icelake-SP because of the yields (which of course affects the competitveness).

 

[Yotsugi]

Yotsugi, just because you and Charlie says something is so, doesn’t make it so

The harsh reality is exactly what it is.

That isn’t a coping mechanism

Oh you're coping like noone else.
Adorable.

What do you do for a living?

Signing papers.
Very fun.

Pretty obvious they "delayed" (read: it's getting cancelled) Icelake-SP because of the yields (which of course affects the competitveness).

At least they had enough balls to try and get XCC outta the oven, I give them that.

 

[jpiniero]

At least they had enough balls to try and get XCC outta the oven, I give them that.

That's why I brought up chiplets for Saphhire. If it was small enough, even with garbage yield they should still be able to produce a decent amount.

Whether that would be enough to supply the market is another story, unless market share craters.

 

[Yotsugi]

That's why I brought up chiplets for Saphhire

SPR is probably a few big dies bridged together.

 

[jpiniero]

SPR is probably a few big dies bridged together.

That's the mystery. Guess we'll have to see.

I can't see them selling 14 nm server parts against AMD on TSMC 5 nm even with Intel's bribe money.

 

[Dayman1225]

Intel denies the rumours and says

We continue to make great progress on 10nm, and our current roadmap of 10nm products includes desktop.

Intel: Yes, There Will be 10nm Desktop CPUs (Updated)

Intel refutes rumors that it is skipping the 10nm process for desktop processors

www.tomshardware.com

 

[Yotsugi]

Intel denies the rumours and says

Intel: Yes, There Will be 10nm Desktop CPUs (Updated)

Intel refutes rumors that it is skipping the 10nm process for desktop processors

www.tomshardware.com

Wow we totally should trust Intel, for they never ever lied to us right

 

[Glo.]

NUC's with Mobile CPUs are also desktop .

So they can offer the new form factor NUCs with Mobile CPUs on desktop and say they delivered on the promise

 

[jpiniero]

NUC's with Mobile CPUs are also desktop .

So they can offer the new form factor NUCs with Mobile CPUs on desktop and say they delivered on the promise

I was going to say desktop versions of any Atom parts that get released.

Should note that Elkhart Lake does have a IGP included, and while the die has 32 EUs it's being configured to go down as far as 8 EUs.

 

[liahos1]

Lol Yotsugi - your coping mechanism is funny