Intel's Foundry Business Hinges on 14A Node Success

[TheLostSwede]

I think you missed the point. The cross licensing agreement applies to all x86 related items between Intel and AMD. SSE, x64, etc. are all under the agreement to prevent Intel from rewriting the spec to lock out AMD. That's mostly what the Itanic was all about. If Intel could have produced a new server architecture without x86, they could have cornered the server market. And since AMD sank it by creating x64 on top of x86, AMD was forced to surrender it to Intel per the cross license.

In other words, only Intel and AMD can make x64 chips, and the terms of the cross license that apply to a sale of the license still apply as well.

That's not how it works though. If that was true, why do both companies have all these extra bits that the other company can't use, so they make up their own variants of it?
It might apply to the basic instruction set, but it doesn't apply to the entire architecture from either company.
Intel even tried to make this:

Envisioning a Simplified Intel® Architecture

Intel investigated architectural enhancements and modifications for a 64-bit mode-only architecture.

www.intel.com

They also appear to claim they invented x64 on that page, which is a lie.

You forgot about VIA, they still have a license via Winchip/Cyrix, even though Intel bought most of Centaur a couple of years ago. That license is now being used by Zhaoxin, even though they're years behind due to the lack of investment by VIA.

Then you have Vortex86 that also has an x86 license, but not an x64 license.

So not, it's not only Intel and AMD.

Wasn't it supposed to hinge on 18A? I guess it's hinging on 14A now?

New CEO, new rules.

 

[Chomiq]

It hinges so heavily that they just officially cancelled FAB projects in Poland and Germany. Intel just has some sort of identity crysis rn.

 

[TheLostSwede]

It hinges so heavily that they just officially cancelled FAB projects in Poland and Germany.

Only Germany was supposed to get a chip making plant though, the Poland thing was a testing and assembly plant.
I guess Greg no longer has a job at Intel...
Must suck having learnt a language that will be of no use to you.

Greg Anderson - Intel Wroclaw West, Lower Silesia, Poland

Get to know Greg Anderson, General Manager of Intel's planned Assembly and Test Facility in Wroclaw West, Lower Silesia Poland.

www.intel.com

 

[io-waiter]

Shareholders money is all that counts. Removing fabs removes the potential edge of profit through integration but it significantly reduces risk. All relying on TSMC and some third rate foundries as backup puts all at the same risk.

Big question is if the US and to some extent the EU will allow themselves to but completely reliant on TSMC, Taiwan is a stones throw from mainland China and China is rather bullish in expanding their own fabs as well as their territory.

Intel is not going to come out of this being the Intel of old, they will be more Broadcom and Qualcomm like.. as the shareholders and the board requested.

 

[watzupken]

As much as I don’t like to admit, I feel Intel’s has passed the point of no return. They have taken their eyes off creating enticing and competitive products, and their cost cutting measures can only take them that far. Losing headcount and limiting creativity is probably going to make matter worst in the longer run. They should have cut loose their foundry business earlier. It’s a great business, but it was clear that they lost the foundry advantage when previous CEOs just took their time to wait for 10nm to “hatch”. To add to the problem, Intel never used to work with external chip designers and from a relationship standpoint, TSMC clearly have the advantage.

The CEO’s statement around canning 14A and future cutting edge node feels more like a desperation warning that they are the only company that can provide cutting edge node as an alternative to TSMC. Samsung already crashed out of the race.

 

[TheLostSwede]

Shareholders money is all that counts. Removing fabs removes the potential edge of profit through integration but it significantly reduces risk. All relying on TSMC and some third rate foundries as backup puts all at the same risk.

Big question is if the US and to some extent the EU will allow themselves to but completely reliant on TSMC, Taiwan is a stones throw from mainland China and China is rather bullish in expanding their own fabs as well as their territory.

Intel is not going to come out of this being the Intel of old, they will be more Broadcom and Qualcomm like.. as the shareholders and the board requested.

The whole point of the planned Intel fab in Germany was to reduce risks of a shortfall of semiconductors in Europe, but it seems like the EU is going to have to sit down and work out a means of making their own fabs now. Admittedly Intel has fabs in Ireland, of which the most advanced one is on Intel 4 and it's said to move to Intel 3 this year.

It seems like this is par for course for a lot of US companies, which isn't healthy.

As much as I don’t like to admit, I feel Intel’s has passed the point of no return. They have taken their eyes off creating enticing and competitive products, and their cost cutting measures can only take them that far. Losing headcount and limiting creativity is probably going to make matter worst in the longer run. They should have cut loose their foundry business earlier. It’s a great business, but it was clear that they lost the foundry advantage when previous CEOs just took their time to wait for 10nm to “hatch”. To add to the problem, Intel never used to work with external chip designers and from a relationship standpoint, TSMC clearly have the advantage.

Intel was a dinosaur in many ways, but what they're doing now reeks of panic and that is never good. The problem as I see it, is that they've tried to get involved in so many things outside of their core business, that they lost focus. I mean, who thought it was a good idea to buy McAfee for US$7,68 billion? Admittedly they made some money when they sold it, unlike so many of their other business units that they either just closed or sold at a massive loss. There also seemed to be a lot of internal issues with some BUs where they just never delivered a competitive product or even a fully working product in some instances.

The CEO’s statement around canning 14A and future cutting edge node feels more like a desperation warning that they are the only company that can provide cutting edge node as an alternative to TSMC. Samsung already crashed out of the race.

Well, there's a reason why I decided to use a different headline compared to this publication. I guess I interpreted what was said differently, but I guess we'll have to wait and see what happens.

Intel will cancel 14A and following nodes if it can't win a major external customer — move would cede leading-edge nodes to TSMC and Samsung

Then exit the leading-edge process technologies completely.

www.tomshardware.com

 

[TechBuyingHavoc]

Long term, there were only two solutions for Intel. One was to kill or spin off the design business and compete solely on the fab business. Maybe even open up the x86 duopoly and offer third-party firms the ability to design their own x86 chips and make those chips on Intel fabs. This could have been possible perhaps a decade ago but Intel would have never let go of the x86 profit for an uncertain outcome. On a technical level, Intel fabs have been subject too long to the whims and desires of the design divisions. They have spent so much time fulfilling internal demand using Intel design rules, they don't know how to compete on a purely open market.

Two was to kill or spin off the fab business earlier on and cut all the losses into nodes that will never pay off and compete solely on the design business against Arm, Nvidia, etc, etc. In hindsight, this was the more feasible solution but Intel did not want to decouple their fabs from their design business because as long as the fab nodes arrived on time, it was a powerful firewall for their x86 chips vs the rest of the world. The problem is the fab nodes did not arrive on time (and continue to not arrive on time) so Intel x86 chips look more and more inferior. The x86 ISA itself is not a superior option vs ARM or RISC-V or any other modern ISA so Intel would have had to bet that they could also design great chips on other non-x86 platforms.

Finally, all of this is compounded by Intel demanding too high margins from its potential customers back in the late 2000s and early 2010s. If Intel had offered Atom chips at much lower margins (and continued to invest in this product in a serious manner), they could have dominated the phone business. Same with IoT, Intel never took low-margin product segments seriously so they fell further and further behind on volume vs TSMC. Costs could not drop enough, process node learning was too slow, Intel decided to keep margins high and blow money on distractions like McAfee and the rest is history.

 

[Broken Processor]

Thats what they where saying about 18A and the one before that and the one before that. How many nodes they going to burn through before they make one work.

 

[ARF]

I want Intel to succeed, and I hope they do!

Monopolies are how we got into this mess to begin with.

Monopoly is set by the system, Intel and AMD are not competitors. Maybe if Windows is allowed to run on ARM chips, or Ryzen is allowed to run Android, then we will have an open ecosystem and something which resembles competition.

 

[kondamin]

As much as I don’t like to admit, I feel Intel’s has passed the point of no return. They have taken their eyes off creating enticing and competitive products, and their cost cutting measures can only take them that far. Losing headcount and limiting creativity is probably going to make matter worst in the longer run. They should have cut loose their foundry business earlier. It’s a great business, but it was clear that they lost the foundry advantage when previous CEOs just took their time to wait for 10nm to “hatch”. To add to the problem, Intel never used to work with external chip designers and from a relationship standpoint, TSMC clearly have the advantage.

The CEO’s statement around canning 14A and future cutting edge node feels more like a desperation warning that they are the only company that can provide cutting edge node as an alternative to TSMC. Samsung already crashed out of the race.

AMD was in far worse shape than Intel is now, amd pulled trough.
But I do highly doubt that this is a good course the current leadership is taking.

 

[TechBuyingHavoc]

AMD was in far worse shape than Intel is now, amd pulled trough.
But I do highly doubt that this is a good course the current leadership is taking.

AMD was in far worse financial shape but the competitive environment around AMD in 2015-2019 was far less threatening than what Intel is facing now. AMD had nothing to lose and everything to gain with a risky strategy that ultimately paid off, it had no fabs, it had no preexisting business that would be threatened by new moves, and its main competitor in Intel was lazy and complacent which allowed AMD to rapidly innovate out of its failure crouch.

Intel is saddled with its fab investments, it is defending a lot of market territory, it still generates a lot of revenues that could be at risk if it messes up its strategy, the competition is razor sharp and paranoid and there is a lot more competitors (Nvidia, AMD, ARM, all the hyperscalers with their own silicon, the many Chinese companies, etc, etc) so Intel can not miss a single step.

 

[SRB151]

I still believe that Itanium was poorly timed, even considering the massive delays and troubled development of the Merced and McKinley architectures. By 2001, most computers were still running Windows 95/98, 32-bit applications in their relative infancy and full compatibility with 16-bit software was still necessary. The whole distribution dynamics and logistics made even Windows Me something that was hard to accept, as well.

One of the chief complaints about that OS is that it was unstable (largely a byproduct of VXD being dropped and almost every WDM device driver being written for Windows 98 SE instead) and that MS-DOS mode was removed, Windows XP was still some time away, and most machines that were in operation for these markets were barely powerful enough to run NT 4, something completely counter-intuitive and very expensive for the average consumer at the time. Dropping an all-new, pure 64-bit ISA in that market was never going to work. Heck, unless you're Apple with its tremendous amount of vertical integration and have zero regard for backwards compatibility, it's a difficult proposition today. With modern compilers and advanced software emulation, I don't believe Itanium would have flopped anywhere near as hard, especially if it could pull its weight in benchmarks and priced well.

x86's patent time is slowly ticking, I believe that most essential patents to make an AMD Clawhammer-compatible (1st gen Athlon 64) should be just about expiring or expired very recently, unfortunately Google is very uncooperative and researching that turns out to some of my own posts on this forum about this subject... it's just an educated guess based on patents lasting between 17 to 20 years.

The latest cross licensing agreement between AMD and Intel was renegotiated in 2009. Every iteration since then (such as the latest AVX) becomes part of the chip and is covered by patent.

That's not how it works though. If that was true, why do both companies have all these extra bits that the other company can't use, so they make up their own variants of it?
It might apply to the basic instruction set, but it doesn't apply to the entire architecture from either company.
Intel even tried to make this:

Envisioning a Simplified Intel® Architecture

Intel investigated architectural enhancements and modifications for a 64-bit mode-only architecture.

www.intel.com

They also appear to claim they invented x64 on that page, which is a lie.

You forgot about VIA, they still have a license via Winchip/Cyrix, even though Intel bought most of Centaur a couple of years ago. That license is now being used by Zhaoxin, even though they're years behind due to the lack of investment by VIA.

Then you have Vortex86 that also has an x86 license, but not an x64 license.

So not, it's not only Intel and AMD.


New CEO, new rules.

Here is a better summary then I can provide.....

AMD could block the sale of Intel due to a cross-licensing agreement

Things get...complicated.

www.tomshardware.com

I don't believe VIA is a party to the last few cross license agreements about the latest extensions. And cross licensing does not mean any new extensions are shared before release, or even required to be included (AVX-512 for one).

 

[Dr. Dro]

The latest cross licensing agreement between AMD and Intel was renegotiated in 2009. Every iteration since then (such as the latest AVX) becomes part of the chip and is covered by patent.


Here is a better summary then I can provide.....

AMD could block the sale of Intel due to a cross-licensing agreement

Things get...complicated.

www.tomshardware.com

I don't believe VIA is a party to the last few cross license agreements about the latest extensions. And cross licensing does not mean any new extensions are shared before release, or even required to be included (AVX-512 for one).

Yeah, the latest instruction sets certainly are still very much patented, but the original AMD64 chips' capabilities should pretty much be getting to their expiry point by now. 130 nm Clawhammer chips came out in 2003, so that's 22 years... assuming 25 years, that'd be by 2028. But I doubt anyone is actually going to try and design an open source x86 chip with these technologies any time soon, nor do I see why that would be useful, it'd end up like ReactOS, "it exists, I guess?"

 

[nguyen]

INTEL: our next node will be competitive, just wait guys

 

[Bomby569]

Intel foundries is now a announcement machine, in fact it became one long ago. Stop announcing things and actually make them

 

[TheLostSwede]

Yeah, the latest instruction sets certainly are still very much patented, but the original AMD64 chips' capabilities should pretty much be getting to their expiry point by now. 130 nm Clawhammer chips came out in 2003, so that's 22 years... assuming 25 years, that'd be by 2028. But I doubt anyone is actually going to try and design an open source x86 chip with these technologies any time soon, nor do I see why that would be useful, it'd end up like ReactOS, "it exists, I guess?"

And just because a patent expires or is opened up, doesn't mean there aren't some "secret sauce" bits that would make it very hard for someone to try and make a chip based on the patents.
Just look at USB4, which Intel so kindly "donated" Thunderbolt 3 for, but somehow assumed everyone knew how Intel does things internally, which is why there have been so many delays when it comes to USB4 chips.

 

[Dr. Dro]

And just because a patent expires or is opened up, doesn't mean there aren't some "secret sauce" bits that would make it very hard for someone to try and make a chip based on the patents.
Just look at USB4, which Intel so kindly "donated" Thunderbolt 3 for, but somehow assumed everyone knew how Intel does things internally, which is why there have been so many delays when it comes to USB4 chips.

I kind of want some development on vintage x86 CPUs though. Being able to create some sort of SoC with a relatively performant Pentium MMX or even Pentium II compatible core, SVGA/hi-color graphics (maybe S3 Trio64 compatible, it's a well documented graphics adapter by now), should facilitate the creation of a proper successor to the Pocket 386, not sure you know of it, it's a computer reminiscent of the old Toshiba Libretto laptops.

I really wanted to get one, but with the outrageously high import taxes currently going on in Brazil (and about to get worse soon), it's just out of the question, a 386 SX/40 is not a very useful system as that CPU is too slow to even play a 128 Kbps MP3 file

 

[TheLostSwede]

I kind of want some development on vintage x86 CPUs though. Being able to create some sort of SoC with a relatively performant Pentium MMX or even Pentium II compatible core, SVGA/hi-color graphics (maybe S3 Trio64 compatible, it's a well documented graphics adapter by now), should facilitate the creation of a proper successor to the Pocket 386, not sure you know of it, it's a computer reminiscent of the old Toshiba Libretto laptops.

I really wanted to get one, but with the outrageously high import taxes currently going on in Brazil (and about to get worse soon), it's just out of the question, a 386 SX/40 is not a very useful system as that CPU is too slow to even play a 128 Kbps MP3 file

These guys have your back. They even have support for semi-modern RAM and most modern interfaces.

Vortex86

The Vortex86 SoC series are 32-bit x86-architecture microprocessors that are compatible with Windows, Linux and most popular 32-bit Real Time Operating Systems.

www.vortex86.com

 

[igormp]

I kind of want some development on vintage x86 CPUs though. Being able to create some sort of SoC with a relatively performant Pentium MMX or even Pentium II compatible core, SVGA/hi-color graphics (maybe S3 Trio64 compatible, it's a well documented graphics adapter by now), should facilitate the creation of a proper successor to the Pocket 386, not sure you know of it, it's a computer reminiscent of the old Toshiba Libretto laptops.

I really wanted to get one, but with the outrageously high import taxes currently going on in Brazil (and about to get worse soon), it's just out of the question, a 386 SX/40 is not a very useful system as that CPU is too slow to even play a 128 Kbps MP3 file

There are a handful of 32-bit x86 cores floating around already. Main issue is you turning those FPGA designs into actual silicon with all required peripherals for your intended use case.

 

[Deorc Mearh]

At the moment I don' t like this CEO at all. It seems to me he has been chosen just for his Asian ethnicity but he is making a lot of inappropriate declarations. I don't want to judge the choices but declarations for the stocks and the financial stability are important and he is magking a lot of communication mistakes that are making the title weak from ill intentioned subjects. By many articles I've been reading from months on other websites I can clearly see a mediatic campaign to terrorize small investors so that they sell off their stocks: in my opinion this campaign has being promoted by those subject who want to acquire Intel's foundries. This CEO is helping them a lot.
By his statements is quite immediate to guess the 18A process is a complete failure - be it true or not - . Before raising the bid with the 14A claim he should have reassured the investors with some pretests like "we prefer to concentrate our efforts on our production as we are very confident our next products will improve substantially and we limit for the moment the investments in new facilities while we are recovering the turnover from the past years".
I'm not English mothertongue so this is just a coarse idea. He looks lost and with no vision, no enthusiasm, no competence and he is dismissing too many activities that has given Intel some prestige with under an apparent PANIC.
I'm so sure Intel is underrated now that I decided to invest over it, thou in the very past I used to detest Intel. They are on the right path in my opinion, they still produce the best cpus for floating point appliations, Arrow Lake mobile cpus are amazing, the price of a 265k is now more than reasonable. They just need enough tranquillity to improve they foundry process and design. The idea of turning CPUS in gpus with many cores is good to me and I appreciate they are developing gpus as well. They should target to a cheaper version of 5070 gpu range in my opinion. No more but not less.

Long term, there were only two solutions for Intel. One was to kill or spin off the design business and compete solely on the fab business. Maybe even open up the x86 duopoly and offer third-party firms the ability to design their own x86 chips and make those chips on Intel fabs. This could have been possible perhaps a decade ago but Intel would have never let go of the x86 profit for an uncertain outcome. On a technical level, Intel fabs have been subject too long to the whims and desires of the design divisions. They have spent so much time fulfilling internal demand using Intel design rules, they don't know how to compete on a purely open market.

Two was to kill or spin off the fab business earlier on and cut all the losses into nodes that will never pay off and compete solely on the design business against Arm, Nvidia, etc, etc. In hindsight, this was the more feasible solution but Intel did not want to decouple their fabs from their design business because as long as the fab nodes arrived on time, it was a powerful firewall for their x86 chips vs the rest of the world. The problem is the fab nodes did not arrive on time (and continue to not arrive on time) so Intel x86 chips look more and more inferior. The x86 ISA itself is not a superior option vs ARM or RISC-V or any other modern ISA so Intel would have had to bet that they could also design great chips on other non-x86 platforms.

Finally, all of this is compounded by Intel demanding too high margins from its potential customers back in the late 2000s and early 2010s. If Intel had offered Atom chips at much lower margins (and continued to invest in this product in a serious manner), they could have dominated the phone business. Same with IoT, Intel never took low-margin product segments seriously so they fell further and further behind on volume vs TSMC. Costs could not drop enough, process node learning was too slow, Intel decided to keep margins high and blow money on distractions like McAfee and the rest is history.

you are not considering the big picture and this involves geopolitics: if ever TSMC undergoes Taiwan invasion, having inside foundries ready to produce for clients might give Intel a huge advantage

 

[Visible Noise]

Thats what they were saying about 18A and the one before that and the one before that. How many nodes they going to burn through before they make one work.

Again, 18A is in mass production today.

By many articles I've been reading from months on other websites I can clearly see a mediatic campaign to terrorize small investors so that they sell off their stocks: in my opinion this campaign has being promoted by those subject who want to acquire Intel's foundries.

Every small investor on the planet could tender every single share they own, and if a single entity bought them all they might get a seat on the BOD. I don’t think you understand how large Intels float is, and how much of it is held by institutional investors.

Institutional investors hold 2.9 billion - yes with a B - shares of Intel. Nobody is trying to play games with individuals - there’s no point.

 

[Broken Processor]

Again, 18A is in mass production today.

Bad yields and profit makes it a failed node.

 

[mrnagant]

What is the low down and Intel chip production history over the last 10 years. I feel like Intel's last successful nm was 14nm, which turned into 14nm+++++++ and eventually got so many +'s that it turned into 10nm. Then Intel started calling their 10nm Intel 7 when TSMC was really hitting home runs. Seems like Intel has been on nothing but the R&D / Production / Cancellation struggle bus post-14nm.

 

[kondamin]

Bad yields and profit makes it a failed node.

Where have you read recent yield numbers? Wccf is not something to quote

 

[phanbuey]

What is the low down and Intel chip production history over the last 10 years. I feel like Intel's last successful nm was 14nm, which turned into 14nm+++++++ and eventually got so many +'s that it turned into 10nm. Then Intel started calling their 10nm Intel 7 when TSMC was really hitting home runs. Seems like Intel has been on nothing but the R&D / Production / Cancellation struggle bus post-14nm.

Facts.

Again, 18A is in mass production today.

So then why isn't Nova Lake on 18A?