[SemiAccurate] Intel kills off the 10nm process

Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
I've seen whats behind it and it certainly isn't a new 14nm Server CPU arch. Its a 4s Server platform named Cedar Isle for Cooper Lake and potentially Icelake

Ah interesting. I've always thought Icelake Server was going to be a paper launch at best because it was so unrealistic to think they could get yield up that quickly. I still think it's cancelled.

Charlie makes it sound like the decision to kill the Real 10 nm was fairly recent. Maybe they could Sapphire Rapids' CPU tile to the Fake 10 nm for release in Early 2021. Would 2 years be enough time?
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Ah interesting. I've always thought Icelake Server was going to be a paper launch at best because it was so unrealistic to think they could get yield up that quickly.

Usually once the first volume run happens, everything happens very very quickly. If they can get U and Y chips out in full force, then the rest will follow much faster than you expect. The U chips are likely quite a decent chunk of their mobile chip sales, so not low volume at all.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Usually once the first volume run happens, everything happens very very quickly. If they can get U and Y chips out in full force, then the rest will follow much faster than you expect. The U chips are likely quite a decent chunk of their mobile chip sales, so not low volume at all.

For 14 nm, it was like a year and a half. Getting the defect rate down to something where a 90-100 mm2 die is doable is one thing, a 350-400 one is another story.

And if Intel is using a completely different "10 nm" node for Icelake client, the question is whether Icelake Server is even built on that.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
For 14 nm, it was like a year and a half. Getting the defect rate down to something where a 90-100 mm2 die is doable is one thing, a 350-400 one is another story.

And if Intel is using a completely different "10 nm" node for Icelake client, the question is whether Icelake Server is even built on that.
IIRC, the defect rate is an exponential as a function of die area, so it's a big challenge to create a die that’s 4X larger with a decent yield.
 
  • Like
Reactions: Arachnotronic

PPB

Golden Member
Jul 5, 2013
1,118
168
106
IMO charlie is baiting us non subscribers here and probably this is just a matter of canning the first iteration of 10nm,which doesnt mean 10nm goes boom altogether.

If we remember 14nm came in a bad shape in anything outside of mobile, broadwell R had painful low clocks and if memory serves right, skylake had to relax the pipeline lenght to regain clocks, then with 14+, 14++ clocks went to what we see today in CFL.

Probably we will see 10nm products, just less dense ones.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
If we remember 14nm came in a bad shape in anything outside of mobile, broadwell R had painful low clocks
Mobile Broadwell was an upgrade over Haswell in sustained performance. Cannon Lake doesn't even have a GPU. To put this in very blunt terms, we're comparing mediocre with retard.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
The density changes will likely not impact their Core line.

I think it will, and possibly quite greatly. It's true that Intel doesn't push the density as hard as they could with Core, but this new process is likely not anywhere near comparable to what Intel got with Cannonlake (which was pretty much half the size). But we won't really know until Icelake ships.

On the old process, I had estimated it to be 100 mm2 (with 4 cores and 64 EU).
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
The original plan for 10nm claimed 2.7x increase in density. They claimed 2.5x increase in density for 14nm, yet we only saw 2x for Core. So even if the new 10nm is 33% larger, you still get 2x density, which is in line with preceding process generations.

They'll scale to 0.5x the size(2x density) the same as in previous generations and in the big picture won't matter.
 
Mar 10, 2006
11,715
2,012
126
I think it will, and possibly quite greatly. It's true that Intel doesn't push the density as hard as they could with Core, but this new process is likely not anywhere near comparable to what Intel got with Cannonlake (which was pretty much half the size). But we won't really know until Icelake ships.

You make these claims with such certainty. Why?
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
The original plan for 10nm claimed 2.7x increase in density. They claimed 2.5x increase in density for 14nm, yet we only saw 2x for Core. So even if the new 10nm is 33% larger, you still get 2x density, which is in line with preceding process generations.

But would they be able to get 2x density out of Core?

I'm not concerned about the density per se, although that would obviously hurt Intel's margins. My real question is what compromises they had to make in order to get it out and whether the process is suitable for anything other than mobile.
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
I think if they can release Icelake mobile 4C/8T version for 5W,15W,25W,45W segment with that IPC increase they will be succesfull and can finetune the manufacturing for larger dies.
What I am worried they are not releasing anything. Just postponing over and over.
WTF made the decision to feed 2 years of trying and now killing it? WTB the material they are smoking
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
I think if they can release Icelake mobile 4C/8T version for 5W,15W,25W,45W segment with that IPC increase they will be succesfull and can finetune the manufacturing for larger dies.

There is an Icelake H 8+1 die that was spotted, but I don't know if that was on the old process or the new one. That one likely won't be released until 2020 I reckon if it does.
 
Last edited:

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
There is an Icelake H 8+1 die that was spotted, but I don't know if that was on the old process or the new one. That one likely won't be released until 2020 I reckon if it does.
I can’t wait until we get more facts on the 10nm node and Icelake. It’s still hard for me to wrap my head around the idea that Intel borked an entire generation of chips, save low power mobile. How in the hell are they going to get Icelake-SP out in time?
 
  • Like
Reactions: ryan20fun

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
How in the hell are they going to get Icelake-SP out in time?

They're not, because they killed the 10 nm process and changed the production lines to produce 14 nm instead.

The Icelake Client that's coming out is on an entirely different process that Intel is still calling 10 nm.
 

Vattila

Senior member
Oct 22, 2004
799
1,351
136
I just don’t know how everyone is so sure of this

I think you mean "some" — even "a few". Investors and markets, as well as a lot of enthusiasts here and elsewhere, seem to be content and assured by Intel's reiterated promise of 10nm products by year end 2019. I bet a lot of those thinks it will be back to normal from there — Intel back in the manufacturing lead, beating TSMC's 7nm, with 10nm Ice Lake in all variants appearing in short order.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
They're not, because they killed the 10 nm process and changed the production lines to produce 14 nm instead.

The Icelake Client that's coming out is on an entirely different process that Intel is still calling 10 nm.

Too much drama. People seem to forget 14nm+ is much less dense compared to the original 14nm, and the 14nm++ may be even less dense. Yet i see nobody objecting to calling 14nm++ a 14nm process.

Nobody remotely informed was expecting the original 10nm process from Intel anymore, the only question is by how much intel will relax this node. Charlie is just kicking in open doors, and many times uses dubious sources that may be financially motivated. It's his business model.

But if TSMC with zero experience in high frequency transistor design appears to have a process ready early next year that is near Intel's 10nm density figures and also near intel's 14nm++ frequencies, I have trouble imagining Intel won't be able to hit similar densities and frequencies half a year later with both more experience and a much much longer development cycle. Charlie appears to have hinted behind the paywall this relaxed node will perform much better compared to the orginal process, which again is akin to 14nm vs 14nm+.

It's even possible that intel fixed it's quad patterning (yield) issues yet the cobalt interconnects were structurally underperforming. A different process with similar density may be unlikely, but would not conflict with Charlie's report given the reported issues Intel faced with the original process.

History is repeating itself. The delay is way longer and public this time around with the original process unusable but people writing off Intel's ability to deliver a functional 10nm+ node should calm down a bit, there's no need for hysteria.

That said, the most surprising aspect of all this is not so much Intel screwing up plan A but not having a plan B in place when the original 10nm failed. It's even more bizarre they seem to have acted on a contingency plan to bridge the gap to a plan B with regards to production allocation way too late.

I'm hoping Intel will be open about the way they solved their issues next year (or God forbid the year after) but given the spotlight they've found themselves in, I am pretty sure they will, once their first 10nm+ products hit the shelves.
 
Last edited:
  • Like
Reactions: Zucker2k