Discussion Intel current and future Lakes & Rapids thread

[511]

No, because Intel 7 capacity was always planned to decrease.
Intel 3/4 was never planned to ramp us expeditiously.
The major difference is that Pat planned for a lot of 20A/18A. You know, the node that was canned, and the other node that is now killing Intel margins and having inadequate yields.

20A/18A shares stuff just like Intel 4/3 does so you can simply do 18A in place of 20A.

Intel themselves are admitting that PTL is, at least partly, being limited by yields. Intel does not have the financial horsepower anymore to brute force volume disregarding yields anymore either, and this would have been even worse under Pat.

Ramping takes time and money and PTL yields are not that bad it's just that cost of ramp is high and they don't have the financial horsepower TSMC has the same issue they just have the financial horsepower. The margin for TSMC mature nodes are higher than the N2 one. It's not a Intel issue it's issue of the fab business you need volume to offset the cost.

Also, Pat himself started many of the fab cancellations/delays because he knew how the progress on node development was going. So even if Pat continued to be the CEO, there's no guarantee we would see much more volume than what we see today.

Pat actually put his hand into too many stuff should have just focus the shells and other stuff in one place and Intel wouldn't be having this issue.

 

[Win2012R2]

The major difference is that Pat planned for a lot of 20A/18A. You know, the node that was canned, and the other node that is now killing Intel margins and having inadequate yields.

It's the same node that failed to yield, which is what got Pat fired, after that they renamed 20A into 18A - the main difference was supposed to be that 18A got external PDK ready, which isn't the case.

Maybe by 18A-P they will get yields sorted - and that will likely be what they hoped 20A/18A was meant to be, so basically a delay that they've decided to hide.

 

[511]

It's the same node that failed to yield, which is what got Pat fired, after that they renamed 20A into 18A - the main difference was supposed to be that 18A got external PDK ready, which isn't the case.

How can you say that without that node being pushed into production if 20A has entered production 18A won't have this many issue cause the issue would have been sorted out by than.

Maybe by 18A-P they will get yields sorted - and that will likely be what they hoped 20A/18A was meant to be, so basically a delay that they've decided to hide.

their physical D0 is already fine it's like 0.1 Ian Cutress said that in one of the podcasts and Zinser said that in last call just few days back they are ahead of their yield targets

 

[Win2012R2]

How can you say that without that node being pushed into production if 20A has entered production 18A won't have this many issue cause the issue would have been sorted out by than.

20A got renamed into 18A and that is in production now - with all the yield issues and lack of external PDK, because it is badly yielding 20A.

their physical D0 is already fine

Intel admits yields are not fine or in corpo weasel language - "not where we would like them to be", so no - yields not fine.

Zinser said that in last call just few days back they are ahead of their yield targets

What if targets were changed? After all it's their targets.

 

[511]

20A got renamed into 18A and that is in production now - with all the yield issues and lack of external PDK, because it is badly yielding 20A.

lmao it ain't true 20A was a minimal node like Intel 4.

Intel admits yields are not fine or in corpo weasel language - "not where we would like them to be", so no - yields not fine.

Do you even know that there are two types of yield Parametric and Physical yield their physical is fine it's parametric that's not up to the mark for them.

What if targets were changed? After all it's their targets.

if you change the targets as many time Intel said they say they are on target18A wouldn't even enter risk production lmao.

 

[Hitman928]

lmao it ain't true 20A was a minimal node like Intel 4.

Do you even know that there are two types of yield Parametric and Physical yield their physical is fine it's D0 that's not up to the mark for them.

if you change the targets as many time Intel said they say they are on target18A wouldn't even enter risk production lmao.

Do you know what D0 is?

 

[511]

Do you know what D0 is?

question is do you?
it's defect density i.e the number of defect per area

edit: I made a big typo lol

 

[Hans Gruber]

Maybe Intel should stay on 18A for 6 years like 14nm in the past? That way they will have plenty of time increase their yields.

 

[511]

Maybe Intel should stay on 18A for 6 years like 14nm in the past? That way they will have plenty of time increase their yields.

Intel said they are going to stick to 18A and it's variant for like quite a while to get their ROI out of it not that it's a problem cause even for like 6 Years we would be at like 10A(even this is optimistic)

 

[Win2012R2]

Do you even know that there are two types of yield Parametric and Physical yield their physical is fine it's parametric that's not up to the mark for them.

Yes I (obviously) know, and one needs to get both working. Now here I suspect they've prioritised physical yields but that came at cost of parametric yields, either way it's bad yields overall.

Maybe Intel should stay on 18A for 6 years like 14nm in the past?

You'll have to get enough 18A capacity in the first place to sustain big company, and if it does not produce top bin chips then profitability will be low, and in any case they'd have to invest a lot into getting more 18A fabs running, my guess they'll be super tempted to just try to get external customers for 14A and put money into that.

 

[511]

Yes I (obviously) know, and one needs to get both working. Now here I suspect they've prioritised physical yields but that came at cost of parametric yields, either way it's bad yields overall.

it's not that bad you are making it out to be

You'll have to get enough 18A capacity in the first place to sustain big company, and if it does not produce top bin chips then profitability will be low, and in any case they'd have to invest a lot into getting more 18A fabs running, my guess they'll be super tempted to just try to get external customers for 14A and put money into that.

Each fab in Arizona(fab 52/62) can do 40K WSPM or 80K WSPM total which is a lots of wafer should be enough

Intel's Fab 52 is bigger and better equipped than TSMC's Arizona facilities — Intel's production volume dwarfs TSMC's operations in the U.S.

Best semiconductor manufacturing facility in the U.S., but not fully ramped yet.

www.tomshardware.com

 

[Win2012R2]

it's not that bad you are making it out to be

We should find out by end of this year, my guess is that server chip craze will bail them out again financially, at least this year.

 

[regen1]

IMO, this is simply good news.
I'd like to say "Get out Lip-Bu, Come back Pat right now".

Pat DID the extreme Cap-Ex.
And it was said "too much", however... oh, oh, oh... Intel is being in the wafer shortage.
Today, it has been clear that the all opponents for PG were the fool.

Frankly LBT-Yeary combo haven't done as well as Intel stockbros would have you believe. LBT has done some good but there's also been hasty decisions, U-turns and opacity(unclear or no roadmaps). Messaging on foundry is fuzzy/confusing.
Some quarters ago 18A-family became a no go for external and then the messaging went in opposite direction very recently. Very vague on 14A as well. Pull-out from Synopsys IP deal seemingly straining the relation. Intel 3 still hasn't ramped as much as it should have, GNR products are relatively scarce after 18+ months from launch/announcement. Mismanaged Foundry and Server priorities about 3 Quarters ago(which lead to Stacy's question on Foundry and Demand mismanagement in last Earnings call).
Communication on Discrete GPUs is "Error 404". Then the controversial acquisition attempt but finally better sense prevailed, the SambaNova deal is much better than full acquisition.
Network Division was tried to be sold off once again(it was tried under Pat and then halted), will give a benefit of doubt here. Double whammy of DMR-SP/DMR-8CH cancellation leaves Enterprise(a huge stronghold for Intel) with no new Intel options, potential market-share loss and lower foundry utilization. Stupid work policies leading to talent-bleed of another level. Then there's more.

It's not his first year at Intel, he was in the Board for 2 years. Hopefully with a new Chairman, the Board including LBT make better decisions and progress.

 

[511]

Frankly LBT-Yeary combo haven't done as well as Intel stockbros would have you believe. LBT has done some good but there's also been hasty decisions, U-turns and opacity(unclear or no roadmaps). Messaging on foundry is fuzzy/confusing.
Some quarters ago 18A-family became a no go for external and then the messaging went in opposite direction very recently. Very vague on 14A as well. Pull-out from Synopsys IP deal seemingly straining the relation. Intel 3 still hasn't ramped as much as it should have, GNR products are relatively scarce after 18+ months from launch/announcement. Mismanaged Foundry and Server priorities about 3 Quarters ago(which lead to Stacy's question on Foundry and Demand mismanagement in last Earnings call).

Yeah

Communication on Discrete GPUs is "Error 404". Then the controversial acquisition attempt but finally better sense prevailed, the SambaNova deal is much better than full acquisition.

Samba nova deal exists due the fact that he is chairman on the samba board.

Network Division was tried to be sold off once again(it was tried under Pat and then halted), will give a benefit of doubt here. Double whammy of DMR-SP/DMR-8CH cancellation leaves Enterprise(a huge stronghold for Intel) with no new Intel options, potential market-share loss and lower foundry utilization. Stupid work policies leading to talent-bleed of another level. Then there's more.

Yeah peak stupidity in these decisions.

We should find out by end of this year, my guess is that server chip craze will bail them out again financially, at least this year.

Have to see but looks like server will grow very much and they don't have the capacity at Intel 7/3 so they are capacity constrained.

 

[Doug S]

Intel's Fab 52 is bigger and better equipped than TSMC's Arizona facilities — Intel's production volume dwarfs TSMC's operations in the U.S.

Best semiconductor manufacturing facility in the U.S., but not fully ramped yet.

www.tomshardware.com

There's an interesting tidbit in that article I think should be highlighted:

There is a caveat regarding Intel's Fab 52 ramp schedule, however. At present, it is ramping up production of Intel's Panther Lake processors using 18A technology, which is still in an early stage of its yield curve. Intel expects 18A yields to reach world-class levels in early 2027. Before then, Intel will not increase CPU production on this node beyond a certain level

That reads to me as 18A is in what TSMC terms risk production, and won't reach what TSMC terms mass production until 2027.

If you read it that way the timelines regarding 14A being in risk production in 2028 and mass production in 2029 don't sound so bad - that's a two gear gap for a new node which is pretty good these days - considering that even mighty TSMC took three years go from N5 to N3B, and three years to go from N3B to N2 (or if you prefer, four years to go from N5 to N3E, and two years to go from N3E to N2)

 

[Khato]

It's pretty clear that the cause for PTL's 5.1GHz max frequency is that the 18A process didn't meet the specifications of the PDK. It's not going to be the only process this year that fails to live up to its promises.

LBT's only strength seems to be his industry connections. I was initially optimistic with his promise to reduce middle management as that is indeed the bane of Intel's existence... but a year later I'm told there's still no meaningful change in that respect. What has changed is a blanket RTO mandate because LBT was disheartened by seeing so many empty cubes every day. (Great example of how it's all about him - apparently there was a 2 hour celebration at the Santa Clara campus on the first day of RTO... where the catering staff was instructed to guard the tables of snacks/desserts until after the CEO's speech that was scheduled an hour into the event.) LBT is also a CEO that was so clueless of American politics that a prime opportunity to push Intel foundry instead turned into a scramble to keep his job.

 

[jpiniero]

In fairness to LBT, the foundry is unfixable.

 

[gdansk]

In fairness to LBT, the foundry is unfixable.

I know the doomers dominate on the web, but they are actually shipping PTL chips which suggests it is "fixable" even if not ideal. I can go buy one today and it's a generally better chip for laptops than what AMD is shipping. It's the bare minimum Intel needs to do with 18A but they are doing it.

Problems exist but it seems they are fixing them much more quickly than they did with 10nm.

 

[Khato]

In fairness to LBT, the foundry is unfixable.

On technical merits alone, yes, it would be quite difficult. And continuing Intel's historical norm of scrapping legacy nodes is not the way to do it. The best way to gain ground on TSMC is by taking away legacy node volume, not miraculously leapfrogging them on the leading edge. Which is why it would be quite difficult to achieve success on technical merits alone. And that's the reason why Pat was spending quite so much of his time working the political side.

 

[511]

That reads to me as 18A is in what TSMC terms risk production, and won't reach what TSMC terms mass production until 2027.

It's not that bad tbh TSMC Risk Production D0 is like 0.4-0.3 ish if you look at TSMCs public data Intel's is 0.1 and we are 1Q away from article their yield must have improved quite a bit now also U9 has always been a meme SKU in terms of volume at the start.

If you read it that way the timelines regarding 14A being in risk production in 2028 and mass production in 2029 don't sound so bad - that's a two gear gap for a new node which is pretty good these days - considering that even mighty TSMC took three years go from N5 to N3B, and three years to go from N3B to N2 (or if you prefer, four years to go from N5 to N3E, and two years to go from N3E to N2)

It's 27 risk production for Internal though DZ Said in conference 28 is for external if it materializes.

On technical merits alone, yes, it would be quite difficult. And continuing Intel's historical norm of scrapping legacy nodes is not the way to do it. The best way to gain ground on TSMC is by taking away legacy node volume, not miraculously leapfrogging them on the leading edge. Which is why it would be quite difficult to achieve success on technical merits alone. And that's the reason why Pat was spending quite so much of his time working the political side.

IMO the challenges are in difficulties of use of the process by external and yes ramping down Intel 7 and ramping up Intel 3/18A should solve many of the cost problems

I know the doomers dominate on the web, but they are actually shipping PTL chips which suggests it is "fixable" even if not ideal. I can go buy one today and it's a generally better chip for laptops than what AMD is shipping. It's the bare minimum Intel needs to do with 18A but they are doing it.

Problems exist but it seems they are fixing them much more quickly than they did with 10nm.

It's not 10nm repeat in any scenario if it was they would have been done for.