I find it kinda funny how Samsung uses TSMC's process more than Nvidia.

Apple's slice of the pie is outrageous. No wonder Apple has had no chip troubles, unlike everybody else. From what I've heard, Apple didn't manage to reserve as much from the production of a microfiber cloth factory, though.

Kaarme:

Apple's slice of the pie is outrageous. No wonder Apple has had no chip troubles, unlike everybody else. From what I've heard, Apple didn't manage to reserve as much from the production of a microfiber cloth factory, though.

Very much so. The world could do with less Apple e-waste.

i seriously doubt this report which goes against the grain of contemporary trade publications. there is no doubt that Apple is the primary customer. what is in doubt is the alleged wafer allocations. just like Apple, AMD has paid (through the nose) for wafer allocation/fab time and so has Qualcomm. the parties who would see their allocations reduced would be 1) Samsung, and 2) Mediatek as both have other options @ node and haven't paid for primacy. Nvidia should be unaffected as they paid for being "late to the gate". the facts of the matter include that Apple, AMD, and Qualcomm paid to develop 7n, 5n, and 3n and they'll get their share of fab time before others. if AMD gets additional fab time with Samsung, that's a good thing as it means sales are well beyond projections. but the premise in this article doesn't address this as it treats it as a zero sum game when it's not.

It'd be a good opportunity for Intel to make money, if they sold cpacity of their plants.

I hope they don't move everything over to Samsung, you don't put all the eggs in the same basket! And as for TSMC preference for Apple, they are a business and as such they will gladly take more money for the same amount of work. Blame the cheep that buy a new iph0ck every year whatever the price they're asked for it.

tommo1982:

It'd be a good opportunity for Intel to make money, if they sold cpacity of their plants.

They too are maxed.

wtf crypto is as big as nvidia & AMD now? and I assume thats just bitcoin-only disposable ASICs, jeez if they keep 10x'ing in value every other year, apple might go into a similar situation vs nation-sized miners

tunejunky:

but the premise in this article doesn't address this as it treats it as a zero sum game when it's not.

In a way, it is. See, designing CPUs for a company/process vs another for fabrication is entirely dependent on the tooling and libraries of that fab. You can't design a CPU to be produced at one fab company (say, TSMC) then have another to also produce the same designs (say, Samsung) because their manufacturing processes, tooling and production lines are entirely different, as are the end-result characteristics of the chip such as trace width, electrical properties, trace pitch, etc. That's why the libraries that chip designers use are based on where they are planning to have that specific design made: because the library helps account for all the characteristics of what the end product might look like coming out of the production line. Then comes the steppings where you iterate on your design based on what came out of the initial run while updating the libraries to say "hey, when we designed this bit to look like this, this is how it comes out" so future designs on that same node are more refined. This is why nvidia is going with Samsung for the consumer Ampere chips and TSMC for the professional chips: they saved the most efficient designs for the HPC sector. tl;dr: You can't move a design from one full node to the next without updating the design, much less move it from one fab company to another without a complete overhaul of it. addendum: I'm not saying AMD ISN'T producing producing a chip to be fabbed at Samsung, maybe they are. What I AM saying is that IF they are, that would pretty much be a new design based whatever version of Zen they choose for it.

tunejunky:

the parties who would see their allocations reduced would be 1) Samsung, and 2) Mediatek as both have other options @ node and haven't paid for primacy. Nvidia should be unaffected as they paid for being "late to the gate". the facts of the matter include that Apple, AMD, and Qualcomm paid to develop 7n, 5n, and 3n and they'll get their share of fab time before others.

where are you getting all this? especially who paid to develop what... because first of all 1) I don't think that's how TSMC is functioning. Yes, a customer might pay for allocation. But TSMC invests their own money & resources into development of major nodes. Letting direct customers' invest their own money would be a bad business, because it would just dilute TSMC's profit. And especially so in a market with never-ending demand. But now I am speculating myself. 2) Which brings me to my 2nd point: Any contractual obligations are hidden behind thousands pages of mutual contracts and are not public. So again how do you know who paid what to develop what? 3) And even so, what's stopping Apple to pay even more to increase its primacy in high-demand market like wafers.

What seems to be forgotten is that Apple has only ~5% of the desktop computer market globally (Mac)--in some countries it isn't even 1%, in others it may reach 10%, but the fact is that Apple as a single company doesn't have near the demand that all the other x86 OEM companies combined put on the supply side. iPhone chips aren't anywhere near state-of-the-art in terms of production processes, etc. Intel is still at 10nm, I gather, in it's recently renamed "process 7" marketing--which the by-comparison huge power demands of Alder Lake compared to Zen 3 illustrate well. Also, I've read more than one article this year that states that AMD has exceeded Apple in total 2021 TSMC bookings--which I am not surprised about in the least. But I've read articles that say that Apple's share of TSMC is bigger than AMD's--but I don't think anyone really knows. Lots of rumor-- little fact, as usual.

waltc3:

iPhone chips aren't anywhere near state-of-the-art in terms of production processes, etc.

What are you talking about? The A15 is 5nm and has been shipping for months. AMD isn't even shipping a 5nm product lol. No one cares what Apple's desktop sales are - the vast majority of these shares for apple are Phones/iPads and Macbooks.

I think this would be a very wise move for AMD to make. Even if it would mean, as David3k points out, that splitting production between two very different fabrication processes would require two equally different families of processors. I say this because, by the time 3nm becomes a thing, TSMC might be "under new management." (If you know what I mean...) And/or, they may have several large holes in their roof, and it's my understanding that isn't conducive to achieving production goals.

waltc3:

What seems to be forgotten is that Apple has only ~5% of the desktop computer market globally (Mac)--in some countries it isn't even 1%, in others it may reach 10%, but the fact is that Apple as a single company doesn't have near the demand that all the other x86 OEM companies combined put on the supply side. iPhone chips aren't anywhere near state-of-the-art in terms of production processes, etc. Intel is still at 10nm, I gather, in it's recently renamed "process 7" marketing--which the by-comparison huge power demands of Alder Lake compared to Zen 3 illustrate well. Also, I've read more than one article this year that states that AMD has exceeded Apple in total 2021 TSMC bookings--which I am not surprised about in the least. But I've read articles that say that Apple's share of TSMC is bigger than AMD's--but I don't think anyone really knows. Lots of rumor-- little fact, as usual.

LOL Wut. There is nothing more state-of-the-art than TSMC's 5nm that is actually shipping and that is what Apple used last year. Sorry but you are not following semis very closely if you believe this.

It's only a good move if Samsung's 3nm is as good as TSMC's. nVidia went for it with the 3000 series but we saw many issues and supply was just one of them. It's just my opinion but Samsumg's 8nm isn't great. When i was waiting for my 3090 i did a lot of reading about the switch to a different foundry and they were saying how efficient it would be and that it was going to be this and that. I think about 50% of what they said was either lies or just ignorance from nVidia. We all saw what a $h1t show the 3xxx series launch was and it's still never recovered from it.

All the jokes about Intel delays and in the end AMD might hire Intel to manufacture their CPUs as well. That would be 180 degree turn from 1980's when AMD was manufacturing CPUs for Intel.

waltc3:

What seems to be forgotten is that Apple has only ~5% of the desktop computer market globally (Mac)--in some countries it isn't even 1%, in others it may reach 10%, but the fact is that Apple as a single company doesn't have near the demand that all the other x86 OEM companies combined put on the supply side. iPhone chips aren't anywhere near state-of-the-art in terms of production processes, etc. Intel is still at 10nm, I gather, in it's recently renamed "process 7" marketing--which the by-comparison huge power demands of Alder Lake compared to Zen 3 illustrate well. Also, I've read more than one article this year that states that AMD has exceeded Apple in total 2021 TSMC bookings--which I am not surprised about in the least. But I've read articles that say that Apple's share of TSMC is bigger than AMD's--but I don't think anyone really knows. Lots of rumor-- little fact, as usual.

Well seems the glofo 's and tsmc's marketing worked miracles . Their 16 & 14 nm are actually in density very very close to their failed 20nm process that nobody used , the main difference is the use of finfet for marketing reason they named it lower . So neither the 7 or 5 or 3 are real nm . Now i have no idea if intel's 10 nm or 7 as they call it is inferior or superior to tsmc's n7+ but indeed their transistor density is really close with intel's 10nm . Nowadays talk of nm is pointless , so in my opinion is better to try and compare en with their density and this by it self is not enough but unless they fab the same thing on both fabs we can not compare directly.

Chip production got overwhelmed when Apple shifted from Intel to TSMC a few years ago. That combined with other sectors (ie, automotive, telecom, industrial, etc) that now require far more than before. Also keep in mind that production space is prioritized according to whoever pays the most for it. And Apple always comes out on top (at the expense of others). Thats why Nvidia had to place orders with Samsung for Ampere. Both Nvidia and AMD are nothing compared to the big boys Apple and Qualcomm who got 75% of TSMCs entire 5nm production. And that is why your next gen GPUs will shock you when you see the price. Apple played a big part in that.

David3k:

In a way, it is. See, designing CPUs for a company/process vs another for fabrication is entirely dependent on the tooling and libraries of that fab. You can't design a CPU to be produced at one fab company (say, TSMC) then have another to also produce the same designs (say, Samsung) because their manufacturing processes, tooling and production lines are entirely different, as are the end-result characteristics of the chip such as trace width, electrical properties, trace pitch, etc. That's why the libraries that chip designers use are based on where they are planning to have that specific design made: because the library helps account for all the characteristics of what the end product might look like coming out of the production line. Then comes the steppings where you iterate on your design based on what came out of the initial run while updating the libraries to say "hey, when we designed this bit to look like this, this is how it comes out" so future designs on that same node are more refined. This is why nvidia is going with Samsung for the consumer Ampere chips and TSMC for the professional chips: they saved the most efficient designs for the HPC sector. tl;dr: You can't move a design from one full node to the next without updating the design, much less move it from one fab company to another without a complete overhaul of it. addendum: I'm not saying AMD ISN'T producing producing a chip to be fabbed at Samsung, maybe they are. What I AM saying is that IF they are, that would pretty much be a new design based whatever version of Zen they choose for it.

As AMD is making both CPU's and GPU's and the GPU's seem to be in more short supply, AMD could keep their CPU's at TSMC and move GPU's to samsung like Nvidia did. Even though Nvidia's earning call seems to confirm that they are making more cards then b4 while still beeing in short supply, AMD could increase their market chare in both segments by splitting manufacturing like this.

Kaarme:

Apple's slice of the pie is outrageous. No wonder Apple has had no chip troubles, unlike everybody else. From what I've heard, Apple didn't manage to reserve as much from the production of a microfiber cloth factory, though.

Is outrageous at first sight but when we consider that it was Apple´s money that allowed to invest in better nodes, allowing them to surpass Intel and become the best foundry in the world, is not that shocking. I´ve already said this several time, but Apple is much more than a customer to TSMC, they are their partners, in a certain and without Apple, TSMC wouldn´t be what it is today. Like it or not, Apple deserves the special treatment it gets from TSMC. The real problem lies in the lack of competition in the foundry business.

tommo1982:

It'd be a good opportunity for Intel to make money, if they sold cpacity of their plants.

So true. But first Intel has to sort out all the mess in their own plants...

H83:

Is outrageous at first sight but when we consider that it was Apple´s money that allowed to invest in better nodes, allowing them to surpass Intel and become the best foundry in the world, is not that shocking. I´ve already said this several time, but Apple is much more than a customer to TSMC, they are their partners, in a certain and without Apple, TSMC wouldn´t be what it is today.

Uhhuh... You are just describing "business as usual". When you buy a new car, it allows the car manufacturer to keep developing future cars, using the profit. If you don't buy a new car, the car manufacturer doesn't have money for anything. Even when you buy cheese, the dairy may be able to develop new milk products, to compete against other dairies. TSMC manufactures products for other companies, for profit, and that profit allows it to keep going. Sometimes technology companies do get money even from governments. If memory serves, the USA will help TSMC build the new fab in some state or another. This aligns with American national interest, despite TSMC being a foreign company, so it's apparently okay. Apple is just one technology company among many. People would still have every bit as many smartphones even if Apple had never entered the market. Apple computers are a niche product. Apple may be TSMC's single most important customer right now with really deep pockets allowing TSMC to invest more in development, but without Apple TSMC would still be roughly what it is today. Many other customers need and depend on the advanced nodes as well.