Question Intel Q2: 7 nm in bad shape

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136

Here's the money shot:

The company's 7nm-based CPU product timing is shifting approximately six months relative to prior expectations.
The primary driver is the yield of Intel's 7nm process, which based on recent data, is now trending approximately
twelve months behind the company's internal target.

Data Center volume up 29% (plus 5% ASP)
Notebook up 9%
Desktop down 14% - not as bad as I had thought

Q3 guidance bad - revenue of 18.2b vs 19.2b in 2019
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
You forgot the other half:

"We now expect to see initial production shipments of our first Intel-based 7nm product, a client CPU in
late 2022 or early 2023... including the holiday refresh window of 2022"

First DC CPU on 7nm launched in "first half of 2023."

It does say Ponte Vecchio is a mix of internal and external process.

But its still terrible news. They can't afford any more volume loss. It was during the heyday of PCs in 2006-2008 where they recognized the need to increase volume on their process. Hence the focus on mobility(I mean Smartphones not Notebooks) and foundry services.

Yet fast forward more than 10 years, and their foundry service went nowhere and they are out of the mobile market while PC volumes have dropped to 2/3rds what it was.

Now you have Apple defecting, and even Intel products moving to TSMC.

When Intel was the absolute leader in volume they were the first to move to the latest I/O and memory standards.

They absolutely cannot afford any more problems and delays with process. Even if the just reported 7nm problem is the only issue and its fixed, they have to execute with 5nm, and 3nm, and so on and so on.

At this point I am half expecting them to move to a fabless model in 5-7 years. It'll happen slowly, and the resistance will be insane. But you cannot have low volume and have the top tier process technology.

3nm and 2nm is bit of uncertainty at this point. I think I saw the article on Digitimes saying 3nm is half node and 2nm is actual full node jump with GAA.

Man, they've all been slightly more than half nodes since "20nm".

20nm was the density boost with 16nm being performance focused. Each being 0.6 generation.

10nm was the ignored process with most vendors going for 7nm.

Same thing with 5/3.
 
  • Like
Reactions: Tlh97 and Vattila

Geranium

Member
Apr 22, 2020
83
101
61
They still have not fully figured out their 10nm so not sure why anyone would think 7nm would come out with flying colors.

I am betting they will have to cut back on density of 7nm like they did the 10nm to get anything out the door in any salable numbers.
Some thinks AMD did the jump without any problem, so Intel can do the same. But forgets than AMD don't manufacture their chip on their fab any more. AMD fabs chip at TSMC and TSMC figured 10nm before went to 7nm unlike Intel.
 
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Intel's Gross Margin has dropped an amazing 6% over the same period last year indicating while they are selling at a high price, lots of discounts and bundling are happening.

Based on how the market is, most of them has to be happening in the server space.

Some thinks AMD did the jump without any problem, so Intel can do the same. But forgets than AMD don't manufacture their chip on their fab any more. AMD fabs chip at TSMC and TSMC figured 10nm before went to 7nm unlike Intel.

What makes you think that AMD did little to no work in moving to 7nm?

A ton of work has to be done in collaboration with the fab in order to move to a new process. In fact AMD described in detail what they had to do to keep the density/power targets.
 

Failnaught

Member
Aug 4, 2008
26
25
91
This is remarkable news. If Intel fabs all their leading edge chips on TSMC (or Samsung) how do they get the volumes they need? TSMC is fully booked out. Fab capacity takes years to build (so if they want to pass TSMC a couple B$ to get in front of the line, they'd better get on that last month). So... how will it work? Do they have the option of selling their fab to TSMC or licensing the TSMC recipe like how GloFo did with IBM?
 

TheGiant

Senior member
Jun 12, 2017
748
353
106

Here's the money shot:



Data Center volume up 29% (plus 5% ASP)
Notebook up 9%
Desktop down 14% - not as bad as I had thought

Q3 guidance bad - revenue of 18.2b vs 19.2b in 2019
can't find the original plan inside that doc, but since Intel+plan+manufacturing is a littlebig meh It doesn't matter
but BK isn't happening for a long time and 7nm can't be his doing or yes?
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
I'd be willing to bet Apple keeps 3nm tied up though. They will have their entire product line on TSMC and they have a lot more money than AMD. And they are in the exact same position as AMD in regards to marketshare. I know the conventional wisdom is that Apple never goes for marketshare, but they have every reason to try if it will keep team red a year behind them.

The thing is though, the Mac doesn't add much volume to Apple's needs. They sell 200 million iPhones a year, and 50 million iPad/ATV/HomePod (yes the homepod has an A* SoC in it, albeit an older one) plus some watches using tiny SoCs. Adding 20 million Macs to that number, even if you assume they need larger/multiple chips in the high end ones, adds maybe 10% silicon area to Apple's TSMC demand.

Adding Apple's Mac demand in no way makes up for TSMC losing their Huawei demand. Though I would guess a lot of that Huawei demand is replaced by Qualcomm demand as people outside China will substitute someone else's phone for Huawei's.
 

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
This is remarkable news. If Intel fabs all their leading edge chips on TSMC (or Samsung) how do they get the volumes they need? TSMC is fully booked out. Fab capacity takes years to build (so if they want to pass TSMC a couple B$ to get in front of the line, they'd better get on that last month). So... how will it work? Do they have the option of selling their fab to TSMC or licensing the TSMC recipe like how GloFo did with IBM?
Probably Samsung. They aren't booked as tight due to their own issues.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Since TSMC is building a fab in Arizona, AMD and Apple could overcome production limitations with the new fab.

No they can't. That fab is targeting 5nm production and ground won't be broken until next year. By the time it comes online TSMC will have been producing at 3nm for at least a year and 2nm production will be imminent.

Plus it is a small fab compared to the megafabs in Taiwan. If they devoted its entire wafer capacity exclusively to producing iPhone SoCs, and it yielded at 100%, it could only satisfy 75% of Apple's yearly demand - and remember that's for the iPhone alone, it doesn't even include other products that use A* SoCs, or the Macs.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Yes, and anyone still wondering why Apple is going their own way on the Mac instead of sticking with Intel should consider that based on what happened with them at 10nm this may be only the first of several 7nm delays.

But the could have also just gone with AMD. So there is more to it than intels performance problems. I suspect since AMD is usually also behind Apple process with it will start to look award if your "high-end" products use in some way "old" tech. No to mention the actual performance differences.

And I agree, this 6 months delay will become another 6 months delay just like with 10nm. I mean 10nm clearly still had serious yield problems (or some other major issues) or else it's not really explainable that icelake server is still not available. And when it will be, it will face-off vs Milan. Gosh when milan releases Intel will be slaughter on pure numbers in the server space. On actual sales, well I have first hand experience how stupid IT departments can be.


Well the article is true in that we shouldn't get too hung up about process names. Intels 10nm is more like TSMC 7nm and TSMC 5nm is more like intel 7nm. So actually comparing on process name can be misleading by itself.

But yeah, Intel certainly isn't ahead.

It doesn't matter
but BK isn't happening for a long time and 7nm can't be his doing or yes?

Of course it's probably still his doing. Such projects takes years from start to end. And even if he wasn't there anymore when 7nm started, all the talent that left and stupid bureaucratic process and power games implemented all will impact a research department for years if not decade(s). See some problems can only be solved by a few very intelligent engineers. No project management or group effort can change that. If you alienate these important people and implement stupid business processes you get what you have at intel now.
 

gorobei

Diamond Member
Jan 7, 2007
3,654
980
136
I believe that I’ve read in multiple locations that TSMC is still very capacity constrained on their leading edge node products. For AMD to get more capacity, it will take a bidding war with Apple and Nvidia for it among others.
tsmc already indicated amd booked a bunch of capacity and tsmc was giving them some "enhanced" node. they are well aware of how intertwined their future is with amd and its markets potential volume (server epyc, consoles, mobile renoir etc)
nvidia doesnt use the newest nodes because of cost and performance. nv dies are typically large and pushing the reticle limit. at those sizes, early yield rates makes each die expensive because of how few fit on a wafer and the higher defect rate at the beginning. early nodes rarely hit the speeds nv needs.
apple and amd can afford to use early nodes because apple is apple and amd is using chiplets so higher defect rates don't matter because the are cramming tons of chiplets on a wafer.

intel faltering on 7nm as well as 10nm suggests they really have lost the critical individuals that solved the process problems back in the good old days. blame intel's corporate culture and lack of promoting people from the engineering departments.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,573
146
So how is Intel's recently (7 months ago) announced effort of going back to their traditional new node every two years cadence looking now?

At this point this is not even funny to me anymore but plain sad.

Ha ha. Though you're right, and part of me is somewhat worried Intel didn't try to reassure this delay wouldn't affect their 5nm.

Also if you think it's sad man you should have heard the call live. Bob Swan sounded almost broken by the end of it.
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
Does anybody remember this comedy goldmine?

82f241c9-7b87-4380-b321-950ec9520ed4.jpg


https://hexus.net/tech/news/industry/132254-moores-law-dead-insists-intels-jim-keller/
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
@NTMBK

Sucks seeing Keller essentially hung out to dry in a presentation like that. It isn't his fault that things went this way. But hey, he took the stage, so . . . meh.

Anyway.

If anyone is surprised by the idea of Meteor Lake coming out in late 2022, well, don't be. That was actually a pretty rosy estimate that any sane person had BEFORE this announcement from Intel. Ponte Vecchio was supposed to be the pipecleaer for Intel's 7nm, permitting client and server product to launch maybe 6-9 months later. Compare this to Vega20 and Zen2 on TSMC 7nm. Same thing, right? So that put Meteor Lake in Q3 2022, give or take. That's assuming a pain-free launch of 7nm. 6 months delay off THAT, while bad, isn't catastrophic for Intel, especially since AMD seems to be dragging their heels moving onwards towards Zen4.

The real problem here is that Intel has quite-candidly told the public that 7nm EUV isn't working for them.

The LAST time Intel was (eventually) forced to mention that they had problems with a node - namely 10nm - Intel suffered a two year delay moving from failed product (Cannonlake) to working product (Ice Lake-U). Well more like 18 month if you consider that Cannonlake was schlepped out the door unceremoniously in Q4 2017 and Ice Lake-U at least officially launched in Q2/Q3 2019.

Anyway, in "Intel does well" alternate reality, Meteor Lake reaches the market in healthy form in Q3 2022 after a successful launch of a limited number of Ponte Vecchio dGPUs using hybrid 7nm + 10nm chiplet design with internal processes only. In this reality, Meteor Lake could be as late as

drumroll please

Q1 2024.

(that's assuming Q3/Q4 2022 sees a BK-like launch of a 7nm Meteor Lake that is roughly similar in quality and volume to Cannonlake).

Yes, that's right! If we assume that Intel is just that b0rked (and why wouldn't we), they won't even be able to sell client + server CPUs based on their 7nm process until four years from now. Consider for a moment that Intel does not presently have any client CPU plans on 10nm after Alder Lake, and does not have any DCG designs in mind for 10nm past either Sapphire Rapids or Granite Rapids (I honestly don't remember if Granite Rapids was supposed to be 7nm, but I think it was).

Intel's 10nm wafer supply is presently limited compared to 14nm. And I think it was the idea not to expand 10nm capacity too much more since it is a flawed node. So the smart plan would have been to limp along with limited 10nm capacity + expanded 14nm capacity until 7nm was ready, and then go whole-hog on that. But now that would appear to be a losing strategy. Intel has a lot of market obligations to cover. Clients demand hardware. And no, I'm not just talking about Ponte Vecchio. There are people that will want to buy Sapphire Rapids and Alder Lake. There is NO WAY that Intel can replace Cascade Lake with Sapphire Rapids and various 14nm client CPUs with Alder Lake-P/-S given their current 10nm wafer restrictions. On top of this, even if Intel does start expanding 10nm aggressively, they'll have to pull a Kaby/Coffee/Comet-like rehash strategy for their 10nm designs until 7nm is finally ready for some kind of a coherant launch in late 2023/early 2024.

So Intel is now many different kinds of screwed.

Sure, you can assume that Q3 2022 may be another Cannonlake moment, but that they can pull together 7nm faster than they "pulled together" 10nm for Ice Lake-U (never mind that 10nm continued to be flawed after that, but oh well). Maybe Intel gets a full suite of 7nm stuff out before Q1 2024. Q4 2023 maybe? Does it make much difference?

Intel had one shot, and that was to leapfrog 10nm for 7nm and hit the market hard with all those amazing designs we keep hearing about that they've had behind closed doors for so long. Meteor Lake was basically it. Meteor Lake in Q3 2022 would have been great. Q1 2023? Tolerable, I guess, especially if they could then iterate yearly and release something even better in Q1 2024. But a gimped Meteor Lake in Q3/Q4 2022 (remember: it's client-only, NOT for DCG) followed by God-only-knows what? To me, it stinks of Cannonlake Part 2, which is something Intel can not survive, not as a dominant market player and IDM.

Heads will roll. I think Swan won't last much longer.

Intel is sinking fast in pre-market trading.
 

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
Heads will roll. I think Swan won't last much longer.
I hope for everyone's sake they do, because without a shakeup, there is incredible resistance to change and improvement within huge organizations.
It usually needs an outsider with no relation to the employees whatsover to be given a mandate to cleanup the mess. Then when the job is done prepare the rank and file for a new reoganization and new way of working. Lots of veterans provide consultations like these and they charge top dollars for it.
Easier said than done though. But the way things are, it is bound to happen.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I hope for everyone's sake they do, because without a shakeup, there is incredible resistance to change and improvement within huge organizations.

BK's ouster was supposed to be just that. 7nm was supposed to be the savior. I think they're beyond the point of organizational restructuring.
 

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
Tens of thousands of high-tech manufacturing jobs are at risk which will further deindustrialize America. Silicon valley and their high horse righteous political types can get lost because it ain't productive to keep vouching for more employee rights or benefits compared to the far more competitive East Asia ...

Hopefully they'll get used to the lower social status and adjust their political views wisely ... :sunglasses:
 

gorobei

Diamond Member
Jan 7, 2007
3,654
980
136
Tens of thousands of high-tech manufacturing jobs are at risk which will further deindustrialize America. Silicon valley and their high horse righteous political types can get lost because it ain't productive to keep vouching for more employee rights or benefits compared to the far more competitive East Asia ...

Hopefully they'll get used to the lower social status and adjust their political views wisely ... :sunglasses:
politics had very little to do with intel's current failings. this is fundamentally a result of management and corporate culture.

intel used outright illegal practices to get to the top (ip theft, kickbacks, anti competetive, etc) which led to them taking market share from other fabs like amd, texas instruments, national semi, and other US companies. their lobbying arm managed to convince congress to overlook these crimes under the premise that intel was a vital domestic resource even though closer looks at the financials eventual disproved this. at the same time intel was building fabs in other countries.

if intel hadnt reached near monopoly position then the other domestic foundries would still be alive and financially viable and US semiconductor industry wouldnt be threatened by the failing of one company. similar to how bank consolidation has made our finance sector a house of cards ready to fall. lack of diversity and competition is what kills, and intel is to blame for killing its competitors by any means necessary.

if intel had focused on process and products, they would have had the brain drain that led to them losing their lead in manufacturing. if rumors from ex-employees are to be believed a fair share of the high salaried old guard are just milking the cow before it goes tits up (look up green vs blue badge at intel).
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
Tens of thousands of high-tech manufacturing jobs are at risk which will further deindustrialize America. Silicon valley and their high horse righteous political types can get lost because it ain't productive to keep vouching for more employee rights or benefits compared to the far more competitive East Asia ...

Hopefully they'll get used to the lower social status and adjust their political views wisely ... :sunglasses:

Screw the race to the bottom. Put tariffs on goods inversely proportional to the workers' rights in the country of origin.
 

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
I guess that the most likely scenario, if this continues, from the enthusiast point of view is that AMD and Intel swap roles. For the young, you might be getting a live lesson in how a giant dies. learn well.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
I believe that I’ve read in multiple locations that TSMC is still very capacity constrained on their leading edge node products. For AMD to get more capacity, it will take a bidding war with Apple and Nvidia for it among others.
AMD's just starting to manufacture new products on 7nm+ which is already not leading edge, since that's 5nm.