Discussion Intel leading customer for TSMC 3nm?

[Failnaught]

There is news that Intel is the co-leading customer on the TSMC 3nm node, with Apple. Did they manage to outflank AMD on process technology? What are they going to make on this node?

https://asia.nikkei.com/Business/Te...become-first-to-adopt-TSMC-s-latest-chip-tech

 

[Doug S]

^ ^ ^ ^ ^

Other than that, the wccf article states "Nikkei's sources also believe that the overall chip volume TSMC has dedicated to Intel is higher than that dedicated to Apple, resulting in the smartphone maker delaying adopting the 3nm process for its next year's smartphone upgrade." - which sounds really counter-intuitive for me.

TSMC has been consistently saying N3 would ship in volume in "H2 2022". If it comes too late to be used in the fall 2022 iPhones, Apple might not use it in a product until the iPad Pro the following spring. They would never delay an iPhone launch to get a slightly better SoC. Maybe there would be some N3 Macs in that six month span as well, but Apple's real wafer volume begins and ends with iPhone, so no iPhones on N3 (or possibly N3P by that time?) until fall 2023 means TSMC would end up with a lot more unclaimed initial N3 wafers than was the case with N5.

I'm still not 100% sold on that TSMC would allow their process cadence to slip out of step with that of their biggest and most important customer, but the claim in that article does fit with an "H2 2022" that comes too late for Apple to use in next year's iPhone. Sometimes reality doesn't work out the way you plan, and if TSMC found they would be unable to ship N3 wafers in time to meet Apple's schedule, both companies would need to adjust their plans to that new reality. If so, that became Intel's good luck.

As for why Intel is getting those wafers and not AMD, I assume Intel was waving around wads of cash willing to prepay two or three years in advance for early access to N3 wafers and AMD was not. They probably assumed they'd get access to N3 after Apple's allocations, but if the schedule didn't allow Apple to grab all the initial N3 production like they did with N5 then Intel got a better deal than they planned. Despite AMD's recent success, they don't have the huge piles of cash on hand required for buying wafers years in advance of the process even existing like Intel and Apple.

 

[Failnaught]

Intel would get crushed in a spending war with Apple if push came to shove.

I feel like Intel has better place to put their money than getting into a spending war with Apple, of all people. Like a serious stock buyback boost to their R&D labs.

 

[JasonLD]

As for why Intel is getting those wafers and not AMD, I assume Intel was waving around wads of cash willing to prepay two or three years in advance for early access to N3 wafers and AMD was not. They probably assumed they'd get access to N3 after Apple's allocations, but if the schedule didn't allow Apple to grab all the initial N3 production like they did with N5 then Intel got a better deal than they planned. Despite AMD's recent success, they don't have the huge piles of cash on hand required for buying wafers years in advance of the process even existing like Intel and Apple.

AMD most likely doesn't have any product planned for N3 in 2023, since they will be transitioning to 5nm in 2022 and most likely they will stay one more generation on 4/5nm before moving to 3nm.

 

[Kenmitch]

But apple has never cared about ipad as the leading product. Ipads have been rather token amount compared to iphones which have always been the first to a node.

Apple like other companies needs to adjust to the times. Not too long ago you never ever saw Apples products on sale....Jump to current times and you see them on sale all the time.

Then there's things like this....

Apple quietly buys Roku remote button for its struggling Apple TV+ app

Apple TV+ is scrambling for new subscribers — and it’s quietly looking to an unlikely rival for help.

nypost.com

Apple's war chest is much much bigger than Intel's. Additionally, it is extremely unlikely that TSMC would allow Intel to reserve so much capacity as to block long time customers without a long term purchasing agreement from Intel which would very quickly drain that war chest and/or pull money from Intel's own R&D efforts effectively killing Intel's foundry future.

I can understand the temptation, but I don't understand why Intel would want to risk it in such a volatile geographical location. It seams like it would be better to do a joint venture with TSMC and build a new fab on US soil. Maybe it would be easier to just convert a existing fab?

 

[trivik12]

This is no surprise. Pat mentioned it in his very first press conference that he intends to deepen partnership with external foundries. Goal is to reduce risk in execution. He will not want the situation Intel has been in since 10nm delays(from 2016 onwards). I am sure Intel will have a long term supply agreement with TSMC for getting access to leading nodes. Intel will not do any agreement without access leading nodes.

Still this should have little relevance to AMD. AMD is not 1st to leading nodes at TSMC either. Apple is using 5nm since last year and Zen 4 is planned for late 2022. At this pace AMD will be using TSMC 3nm in 2024 and I am sure TSMC will make capacity for them. TSMC will have capacity for Apple as well. Issues seen today are due to shortage of materials that is affecting everyone. By 2023 it will be resolved.

NVidia is mostly using Samsung right. Still if they want they can get access to any note at TSMC. I think Jenson wants to keep costs in check. TSMC leading node will be expensive for sure. Intel will increase prices or lower Gross Margin(may be bit of both).

 

[lobz]

TSMC has been consistently saying N3 would ship in volume in "H2 2022".

That alone would make Intel being a co-lead false.

 

[lobz]

I should point out that 3 nm ramping H2 22 is far too late for the 22 iPhone. Wafer availability wouldn't change that. The timing might work better for an iPad product released in 1H 23.

Apple usually has TSMC start in like February or March for instance.

And when did an iPad use a smaller node half a year before the iPhones? Legit question, I'm waaaaay to lazy to research

 

[Doug S]

That alone would make Intel being a co-lead false.

Why does that make it false? If by "co-load" they mean that all the N3 wafers for the 6-12 months will go to either Apple or Intel, then that's true whether or not N3 is ready in time for the iPhone "14". The only difference is who gets the majority.

 

[JustMe21]

I speculate that Intel is going to use TSMC's 3nm for their HPC stuff.

 

[Schmide]

I'm betting this is more intel trying to show they can play nice and get a chunk of that 52b the gov is supposedly throwing at az. Intel has all this production capability but have never made it generally available. (to rivals) ATM I would say it's a hard sell to steer funding in their direction if it only benefits one company. Whereas funding TSMC for az floats a bunch of companies. Deep guess - TSMC is slated to get that funding and intel knows it.

 

[ThatBuzzkiller]

It should be pretty obvious to anyone to see what Intel is after ...

Intel can't trust their own fabs to deliver and Pat Gelsinger wasn't going to change that one bit like many here thought so he needed a real backup plan in case Intel manages to goof up once more ...

TSMC has virtually no reason to deny service to Intel as long as they're willing to pay a high premium to get access to the latest process technology. Intel paying TSMC just makes it easier for them to defund Intel's fabs in the near future. What TSMC gets in this transaction is loads of money to provide for their indirect competitor. What Intel gets is access to TSMCs wafers but their fabs become temporarily redundant at least in the short term so this is not a one way victory for Intel as some here are suggesting ...

 

[TheELF]

Here is the quote just so everyone at least knows what he said:

Both Meteor Lake and Granite Rapids will have compute tiles built on Intel 7nm. And the strength of our new IDM 2.0 model (integrated device manufacturing) means we can strategically use the breadth of our ecosystem to our advantage. For the 2023 roadmap we will also leverage our relationship with TSMC to deliver additional leadership CPU products for our client and data center customers. This is the power of our new IDM 2.0 model combined with a modular approach to design and Intel’s industry leading packaging technologies.

I have no idea what "other leadership CPU products" are but it seems like it's not Meteor Lake or Granite Rapids ported over based upon that quote. So a new architecture they haven't revealed yet that could theoretically outdo their announced chips on Intel's own process? Or maybe an older architecture tweaked and ported from 10 nm to TSMC 7nm or 5 nm for volume help? It's a pretty ambiguous quote which is why people didn't pay too much attention to it. I have my doubts that Intel will be on 3 nm in 2023 with the type of volume this rumor is indicating but we'll see.

Isn't it pretty obvious with this talk about IDM?
CPU cores will still be by intel but they can easily package tiles made by other companies to save FAB space/produce more valuable stuff. All the 3nm intel gets from tsmc could be used for cache alone, or hbm/igpu, i/o or whatever.

 

[naukkis]

Isn't it pretty obvious with this talk about IDM?
CPU cores will still be by intel but they can easily package tiles made by other companies to save FAB space/produce more valuable stuff. All the 3nm intel gets from tsmc could be used for cache alone, or hbm/igpu, i/o or whatever.

Gelsinger revealed that their chip divisions are free to use whatever is the best suited process for them. So to produce cores with Intel process and using TSMC 3nm for something other can only happen if Intel process is the better one against TSMC 3nm. Those were main points of Gelsinger, Intel isn't any more relying only to it's internal process, their division can use whatever process bring best performance for them.

And Intel's designs today are limited by Intel process they use, that should be perfectly clear to everybody.

 

[IntelUser2000]

And Intel's designs today are limited by Intel process they use, that should be perfectly clear to everybody.

Either Gelsinger is blowing smoke up his ass

OR

He intends to light a fire underneath the process team by having it compete with TSMC and other foundries. Of course he will do what's necessary so the team has the manpower and resources to potentially become a leader again. Announcing many billions of dollars on foundry service and completely give up being a process leader makes no sense.

And Intel's designs today are limited by Intel process they use, that should be perfectly clear to everybody.

And having a core that's behind in both density and power efficiency.

Their deficiency isn't on the process alone. The cores need a serious lesson on downsizing and efficiency too.

 

[DrMrLordX]

MONEY. Bucket loads of it.

I don't see the amounts of money Intel can realistically throw at them being sufficient to be worth TSMC's while. Unless Intel plans to take a loss on every chip they produce at TSMC. There would have to be something else in there causing Intel to effectively become a permanent vassal to TSMC.

Why miss out on the chance to make a buck from your competitor?

If TSMC does nothing to help Intel, they will effectively cede multiple generations of server and workstation sales to competitors. Intel's long-term play is to cooperate with IBM on a 2nm process to be rolled out in ~2024. If their revenue streams are devastated and their market share reduced to scrap in that time period, it may not be possible for Intel to realistically get back in the game by that point. TSMC certainly must be aware that any help they give Intel buys Intel some much-needed breathing room until their 2nm process with IBM can bear fruit.

They have done no such thing. Intel 7nm and TSMC 3nm are within spitting distance of each other (spec wise).

Methinks you overestimate 7nm, but since we haven't seen it in action, there's no real way to know yet.

Essentially they would be paying TSMC a boatload of money to commit suicide faster.

If Intel can still turn a profit on TSMC 3nm, then it buys them some time. Otherwise Intel will be stuck riding (mostly) 10SFE/ESF/whatever until 2024. Which won't be pretty.

 

[andermans]

AMD most likely doesn't have any product planned for N3 in 2023, since they will be transitioning to 5nm in 2022 and most likely they will stay one more generation on 4/5nm before moving to 3nm.

I would guess Zen5 is actually 3nm in 2023. Current rumors/leaks (https://videocardz.com/newz/amd-3nm...rix-point-rumored-to-feature-big-little-cores) are that Strix Point will be based on Zen5, be on 3nm and launch in 2024. If this is like all previous mobile launches this will likely be roughly at CES 2024 with a desktop launch before that (i.e. in 2023). So I'm actually slightly hopeful that AMD is closing the gap a bit between when a node comes online and when they start using it.

 

[jpiniero]

I don't see the amounts of money Intel can realistically throw at them being sufficient to be worth TSMC's while. Unless Intel plans to take a loss on every chip they produce at TSMC. There would have to be something else in there causing Intel to effectively become a permanent vassal to TSMC.

Maybe not necessarily the price of the wafer but the size of the order.

 

[dmens]

If Intel can still turn a profit on TSMC 3nm, then it buys them some time. Otherwise Intel will be stuck riding (mostly) 10SFE/ESF/whatever until 2024. Which won't be pretty.

Not just profit, but big fat margins since Intel is always run by beancounters first and foremost. And good luck with that, since TSMC 5nm and 3nm is when the cost/transistor curve actually inverted. That is the main reason I don't believe this... Intel can't possibly be *that* dumb. Right? They are going to pay a huge amount of money to TSMC for lead production which they will fritter away with their design and validation deficiencies so that by the time the part is actually released it won't even be that far ahead of AMD? Have fun explaining that to shareholders.

 

[jpiniero]

Not just profit, but big fat margins since Intel is always run by beancounters first and foremost. And good luck with that, since TSMC 5nm and 3nm is when the cost/transistor curve actually inverted. That is the main reason I don't believe this... Intel can't possibly be *that* dumb. Right? They are going to pay a huge amount of money to TSMC for lead production which they will fritter away with their design and validation deficiencies so that by the time the part is actually released it won't even be that far ahead of AMD? Have fun explaining that to shareholders.

The alternative is more 10 nm which can't sound too appealing either, esp for servers.

I think they were ready to give up on 7 nm pretty much when they got the Biden Bux idea.

 

[naukkis]

Not just profit, but big fat margins since Intel is always run by beancounters first and foremost. And good luck with that, since TSMC 5nm and 3nm is when the cost/transistor curve actually inverted. That is the main reason I don't believe this... Intel can't possibly be *that* dumb. Right? They are going to pay a huge amount of money to TSMC for lead production which they will fritter away with their design and validation deficiencies so that by the time the part is actually released it won't even be that far ahead of AMD? Have fun explaining that to shareholders.

I think it would be much more attractive to be little bit ahead from rivals than being couple nodes behind. Intel doesn't have benefit of choice, they either quit from server/notebook prime business or go to TSMC, and shareholders should also acknowledge the situation.

 

[ThatBuzzkiller]

Also if TSMC were going to decline service to Intel then the contract would've been given to Samsung who would've taken up their offer regardless which was certainly going to be better than potentially having no viable options with Intel itself. Intel can't even trust itself to manufacture their own designs which purely goes to show how truly bad they have it ...

TSMC not taking up their big offer would've argually been misstep for them since their direct competitor Samsung with far more resources would've came out better funded which might've eventually made it easier for them to catch up ...

 

[naukkis]

And some still don't get that Intel is in middle of catastrophic situation now. They barely have competitive server platform against TSMC 7nm rivals which they don't even be able to produce at needed quantity. Upcoming products are late and are so bad with their specs that it isn't even funny anymore. Their server sales are declining quarter after quarter. And their notebook/desktop sales are only going well because TSMC can't produce more rival's products - that situation will not last forever.

Question is are they actually even trying to make their 7nm anymore - if they are why aren't they ordering more EUV machines? It sure does look like they just gave up with 7nm. For shareholders they will be producing tiniest Risc-V cpu they could find - so they can say that their 7nm is up and running.

 

[TheELF]

Their server sales are declining quarter after quarter.

Annual Reports

www.intc.com

2016 26.5 bil data centric that's before ZEN came out.
2017 28.8 bil
2018 33.8 bil
2019 34.8 bil
2020 37.8 bil

 

[dmens]

Annual Reports

www.intc.com

2016 26.5 bil data centric that's before ZEN came out.
2017 28.8 bil
2018 33.8 bil
2019 34.8 bil
2020 37.8 bil

Did you miss the news of the collapse of Intel's datacenter net revenues the last two quarters?

Intel stock falls despite earnings beat, as data-center sales slump more than 20%

Intel Corp. shares fell in the extended session Thursday despite a big earnings beat and a raised annual forecast, as a large dip in data-center sales was...

www.google.com

 

[maddie]

Annual Reports

www.intc.com

2016 26.5 bil data centric that's before ZEN came out.
2017 28.8 bil
2018 33.8 bil
2019 34.8 bil
2020 37.8 bil

Did you miss the news of the collapse of Intel's datacenter net revenues the last two quarters?

Intel stock falls despite earnings beat, as data-center sales slump more than 20%

Intel Corp. shares fell in the extended session Thursday despite a big earnings beat and a raised annual forecast, as a large dip in data-center sales was...

www.google.com

I remember it being stated that the recent increases in Intel server volumes were related to the security problems of late. It was simpler for customers to add hardware rather than change systems with all the associated validation issues. In other words, the security issues with mainly Intel hardware counterintuitively led to more sales by Intel. That period seems to be ending, as the following quote in the article by dmens states.

On a call with analysts, Intel Chief Executive Pat Gelsinger, in his first earnings report since officially taking over, noted that Intel had just recently launched its new generation of Xeon server chips, codenamed “Ice Lake,” and that the industry is just beginning to emerge from a digestion phase from data centers and that Intel is “starting to see signs that they want to start the next build phase in their cloud.”

The new era begins.