News Intel outlines a plan to get back in the game — pause fab projects in Europe, make the foundry unit an independent subsidiary, and streamline the x...

[TheSecondPower]

Processor designs are insanely complex. In my world, in software, if I have the choice to borrow from existing code, I often pass because it's faster to build a new thing than to understand another thing and borrow from it.

So, then you have absolutely no idea how hard it is for someone to reverse engineer IP from the masks that actually reach the fab? I'll give you a clue: it's a lot harder than just disassembling some code you get in binary form.

I'm not sure why you're making my point for me (about stolen designs being difficult to make use of) while telling me that I have absolutely no idea about it.

As for Global Foundries, I can't say I ever looked through the accounting books but at the time of the split, AMD with foundries and AMD plus Global Foundries were the same assets, same employees, and same products. I don't see where dividing adds value. Maybe if it could convince foundry customers that their designs will be secret, but as I said I don't think that's a huge factor. I think the biggest reason for the split is that investors were willing to put more money into AMD and Global Foundries separately than together. But in my mind Intel is worth more than the combined value of separate entities Intel Design and Intel Foundries.

 

[bit_user]

I would also actually applaud them for having a more cautious and business-responsible approach to jumping onto the AI bandwagon, although this is also showing as a slight weakness as they aren't reaping the cash benefits before the crash happens while AMD and nVidia harness the lucrative opportunity.

Are you talking about Intel, here? Intel has been trying to get on the AI bandwagon since they bought Nervana and Movidius, like 7 years ago! As for "business-responsible", they subsequently wrote off their multi-Billion investment in Nervana and subsequently bought Habana Labs (for $2B, IIRC), in order to close the gap vs. Nvidia a little bit faster.

As if this wasn't enough, they tacked on AMX to every P-core in their server processors, from Sapphire Rapids, onwards. That hurt the competitive position of their server CPUs for non-AI workloads, by chewing up a non-trivial amount of die area (from looking at die shots, maybe 10%?), and yet it's still not fast enough to make them truly competitive vs. GPUs or dedicated AI processors. Also, due to supporting only low-precision arithmetic, it's really not useful for much besides AI.

The final bet they made on AI was in their datacenter GPU product portfolio, where Ponte Vecchio (and its planned successors) leaned heavily on AI as one of the primary applications.

Intel made huge software investments, as well, such as with their OpenVINO AI framework. Also, supporting tools and AI models. I'm sure it factored prominently into their oneAPI investments, as well.

In summary, there's been nothing cautious or terribly business-responsible about Intel's approach to the AI bandwagon. The fact that they're not leading that parade is certainly not for lack of trying!

 

[thestryker]

But Intel only designs CPUs, not laptops or phones or servers.

I'd missed this sentence when reading through the post, but Intel definitely makes laptops (design wise, and sold some retail as recently as ADL) and has specific requirements for several of their branding programs. It's one of the biggest competitive advantages Intel has had in the laptop space over the years and was a driver for AMD's Advantage program (though I don't think this has really expanded much). I've found it quite ironic given that Intel approaches the desktop in a wild west anything goes fashion and haven't really tried to replicate the successful laptop branding.

edit: forgot to mention the server business that they very much had for many years when re-writing this post
https://www.servethehome.com/breaking-intel-exiting-the-server-business-selling-to-mitac/

As for Global Foundries, I can't say I ever looked through the accounting books but at the time of the split, AMD with foundries and AMD plus Global Foundries were the same assets, same employees, and same products. I don't see where dividing adds value. Maybe if it could convince foundry customers that their designs will be secret, but as I said I don't think that's a huge factor. I think the biggest reason for the split is that investors were willing to put more money into AMD and Global Foundries separately than together. But in my mind Intel is worth more than the combined value of separate entities Intel Design and Intel Foundries.

It's actually pretty simple at the end of the day: AMD needed money and spinning off GloFo gave them that. People tend to forget that it wasn't the Bulldozer failure that put them in this position but rather node costs, Intel's illegal OEM dealings and Core architecture launching.

https://www.anandtech.com/show/2635

 

[TheSecondPower]

AMD needed money and spinning off GloFo gave them that.

But how? If the split didn't create something of value, then did AMD essentially move failing aspects of the company to the Global Foundries half before spinning it off? If so, that's not a net win for all investors invloved.

 

[thestryker]

But how? If the split didn't create something of value, then did AMD essentially move failing aspects of the company to the Global Foundries half before spinning it off? If so, that's not a net win for all investors invloved.

Try reading the article I linked as it quite literally explains what was going on, why they spun it off and what the benefit would be.

 

[TheSecondPower]

Try reading the article I linked as it quite literally explains what was going on, why they spun it off and what the benefit would be.

In short, the article says that investor money was what made the split a good idea. It doesn't explain why investors only became interested during the split, but my guess is because they got a great deal of control over Global Foundries. That control didn't make Global Foundries successful, although the money did keep it alive for a long time.

 

[bit_user]

In short, the article says that investor money was what made the split a good idea. It doesn't explain why investors only became interested during the split,

Because an independent fab has more customers than just AMD.

For the design side of AMD, the benefit of being fabless is not only that it doesn't have a capital-intensive manufacturing operation watering down its profits, but also that it can use other fabs, which they did to great effect.

 

[TheSecondPower]

Because an independent fab has more customers than just AMD.

For the design side of AMD, the benefit of being fabless is not only that it doesn't have a capital-intensive manufacturing operation watering down its profits, but also that it can use other fabs, which they did to great effect.

I'm arguing that Intel shouldn't split. Intel already makes chips for 3rd-party customers and already uses an external foundry for many of its consumer chips.

 

[bit_user]

I'm arguing that Intel shouldn't split. Intel already makes chips for 3rd-party customers and already uses an external foundry for many of its consumer chips.

It doesn't really matter what you want, though. Investors like the split, because fabless semiconductor companies have margins more like those of software companies, which is to say big. At a time when the design side of the business is in need of money, a spinoff of the fabs would generate a much-needed cash infusion, much as it did for AMD.

As for your point about consumer processors, that does highlight a way in which Intel's design and fab businesses are still coupled. I highly doubt Intel's designers would be allowed to use TSMC or Samsung for a server CPU, if either of them made more sense.

It's hard to say the fab business would do much better on its own, but at least the story of true independence probably gets some play.

 

[thestryker]

In short, the article says that investor money was what made the split a good idea. It doesn't explain why investors only became interested during the split, but my guess is because they got a great deal of control over Global Foundries. That control didn't make Global Foundries successful, although the money did keep it alive for a long time.

AMD also couldn't get more money for themselves as they'd exhausted options which is the key behind it. Dumping the fabs allowed them to get a cash infusion which was otherwise unavailable.

I'm arguing that Intel shouldn't split. Intel already makes chips for 3rd-party customers and already uses an external foundry for many of its consumer chips.

Anyone who cares about the silicon manufacturing industry (which should be all direct customers of it and end users) shouldn't want Intel to split off the foundry services until it has a lead node wise and enough customers to pay for expansion. The only people who benefit from it spinning off before that time are investors.

 

[joeer77]

This is correct.

This is not correct as there are zero volume production nodes today using GAAFET.

Wrong. Samsung is on 2nd generation GAA node now called SF3. The first generation was called SF3E.

 

[thestryker]

Wrong. Samsung is on 2nd generation GAA node now called SF3. The first generation was called SF3E.

Which complex chips prior to the W1000/Exynos 2500 have been produced using it?

 

[joeer77]

Which complex chips prior to the W1000/Exynos 2500 have been produced using it?

Samsung Galaxy Watch7 is 2nd gen GAA.

 

[joeer77]

Which complex chips prior to the W1000/Exynos 2500 have been produced using it?

Micro BT WhatsMiner M56S crypto miner is first gen GAA.

 

[thestryker]

Micro BT WhatsMiner M56S crypto miner is first gen GAA.

Those are extremely small and basic chips which helps to negate bad yields not to mention Samsung likely wasn't charging a premium.

Samsung Galaxy Watch7 is 2nd gen GAA.

This uses the W1000 which I specifically mentioned and while far more advanced than the mining ASIC still aren't very big. The Exynos 2500 will be the first to fit that.