AMD pushed to the corner by TSMC capacity crunch, ponders tapping Samsung's foundries

AMD CPUs are low margin with intel still getting most all of the Server side. The consumer end just doesn't pay out as much. As far as Nvidia, they have a higher margin because they simply have better RT technology. although AMD is catching up fast. Personally I don't think RT is all that but have to hand it to Nv marketing. Management has little to do with it. You'd be hard pressed to find a company better managed and with a better balance sheet while just becoming profitable. Management has put the R&D pedal to the metal and not letting off.
 
Last edited:
AMD CPUs are low margin with intel still getting most all of the Server side. The consumer end just doesn't pay out as much. As far as Nvidia, they have a higher margin because they simply have better RT technology. although AMD is catching up fast. Personally I don't think RT is all that but have to hand it to Nv marketing. Management has little to do with it. You'd be hard pressed to find a company better managed and with a better balance sheet.
RT is a byproduct of their other processes, the tech powering the 2000, and 3000 series has no competition for datacenter and AI and the money there fuels the toys we get to play with. Their white papers on automation are nothing to scoff at either, they have a solid entry into some upcoming markets that could pay off their ARM purchase (should it finalize) sooner than expected.
 
RT is a byproduct of their other processes, the tech powering the 2000, and 3000 series has no competition for datacenter and AI and the money there fuels the toys we get to play with. Their white papers on automation are nothing to scoff at either, they have a solid entry into some upcoming markets that could pay off their ARM purchase (should it finalize) sooner than expected.
Breaking down Nv 3rd Q
"$1.9 billion in sales in the data center market" That's sales, not profit.
"behind the gaming division"
So let's be clear on where the money comes from to pay for to

"buy ARM from SoftBank for $40 billion"

Of course $11B cash on hand doesn't hurt but obviously it will be financed.
 
So 2 thoughts to this: 1. Yes. Apple takes way too much of TSMC's resources for making iPhones. They need to spare some resources for everyone else. The iPhone isn't exactly the most important piece of technology right about now. I know however as I say that for some people, that is up for debate. 2. AMD getting their own fabs? I like this idea. I like it a lot. Out!
 
AMD getting their own fabs? I like this idea. I like it a lot. Out!
You may like it but it will never happen. AMD can barely keep up R&D with two behemoth competitors. Buillding Cutting edge Fabs for AMD are just too expensive and distracting. They really need to build the behind the scenes infrastructure for CPU support to get Corporate on board. That will raise their margins immensely.
 
Breaking down Nv 3rd Q
"$1.9 billion in sales in the data center market" That's sales, not profit.
"behind the gaming division"
So let's be clear on where the money comes from to pay for to

"buy ARM from SoftBank for $40 billion"

Of course $11B cash on hand doesn't hurt but obviously it will be financed.
You misunderstood, much of the technology that went into the 2000 and 3000 was a direct result of requests from the data centre and AI crowds, which unless they started doing completely custom cores for them much of those features and architectures were going to end up down in the gaming market. The biggest push for Ray Tracing wasn't coming from EA, or Epic, it's been from Disney, WB, and other animation/movie studios which then got Microsoft and NVidia involved which then brought it to games. So much of these things were done for work then they figured out how to use them for play.
 
So 2 thoughts to this: 1. Yes. Apple takes way too much of TSMC's resources for making iPhones. They need to spare some resources for everyone else. The iPhone isn't exactly the most important piece of technology right about now. I know however as I say that for some people, that is up for debate. 2. AMD getting their own fabs? I like this idea. I like it a lot. Out!
Apple pays TSMC a lot of money to get access to those resources, nothing is stopping somebody else from offering TSMC more for their slot. They are a private entity they don't need to hand out resources to anybody, and while the iPhone might not be the most important piece of tech it is one of the most in-demand globally and Apple does its best to keep them available.
AMD getting their own fabs is a terrible idea, AMD does not have the money nor the resources to suddenly develop an all-new manufacturing process that not only beats Intel's 7nm but also TSMC 3nm, which are the processes they would have to compete against if they were to break ground on them today. Which in addition to any R&D costs which would probably be in the 20-30B range, would then have to build said Fabs at a cost of roughly $25B a pop, of which they would need a bare minimum of 2 which would let them meet their current output levels with TSMC. TSMC spends upwards of 4B every year in R&D and an undisclosed amount in maintenance for each of the fabs but an average across them would put them in the 600K range annually. In this hypothetical world they have now spent 70-80Billion over a 5 year period just to switch off TSMC to end up exactly where they were while still on TSMC, there is no series of processes that AMD could then implement to get an ROI on that, and their profit's arent in such a place that they could then continue to maintain the R&D budget needed to keep up with newer node advancements.
 
AMD CPUs are low margin
Retail data shows falling prices (and thus margins) for Intel desktop chips and assumedly same goes for server with their price cuts in the last year.
AMD has very likely overtaken or beginning to overtake Intel for margins.

1614798721387.png
 
It wasn't a value proposition of AMD selling their fabs or not.

My point was that they are now subject to the availability and pricing of someone other than themselves.

Well, you have to consider a few other factors:

1) Even if AMD had their own fabs, we're having shortages of silicon and substrate, so they may not have access to the raw materials to build it themselves anyway, and then they would have the added liability of having fabs.

2) Updating the fabrication process itself is R&D intensive. From what I'm reading, the 7nm node is incredibly fickle, which is why Intel is struggling to launch 10nm after all this time, and GloFo more or less has given up on it. Both those two companies have teams dedicated to this (one of which used to be AMD). For AMD to have not been subject to someone else's fab space, you assume they've also managed to master a 7nm process node on their own, something only TSMC has reliably done at this point (Samsung's close, but a bit behind). Failing that, you have the 14nm+++++ problem that Intel has, which means you have an inferior Ryzen product.

There's a very good reason Nvidia, AMD, Broadcom, Qualcomm, etc, are all fabless chip designers. Fabrication is difficult. You're better off subjecting yourself to someone else's pricing and availability who specialize in it and can give you the very best to work with rather than try to scale up and compete with those specialized manufacturers. That's a worthwhile trade-off for the majority of companies.
 
Well, you have to consider a few other factors:

1) Even if AMD had their own fabs, we're having shortages of silicon and substrate, so they may not have access to the raw materials to build it themselves anyway, and then they would have the added liability of having fabs.

2) Updating the fabrication process itself is R&D intensive. From what I'm reading, the 7nm node is incredibly fickle, which is why Intel is struggling to launch 10nm after all this time, and GloFo more or less has given up on it. Both those two companies have teams dedicated to this (one of which used to be AMD). For AMD to have not been subject to someone else's fab space, you assume they've also managed to master a 7nm process node on their own, something only TSMC has reliably done at this point (Samsung's close, but a bit behind). Failing that, you have the 14nm+++++ problem that Intel has, which means you have an inferior Ryzen product.

There's a very good reason Nvidia, AMD, Broadcom, Qualcomm, etc, are all fabless chip designers. Fabrication is difficult. You're better off subjecting yourself to someone else's pricing and availability who specialize in it and can give you the very best to work with rather than try to scale up and compete with those specialized manufacturers. That's a worthwhile trade-off for the majority of companies.

Shit, tsmc is on 5nm now. They're blowing everyone out of the water.

lol @ AMD getting fabs. How quickly we've forgotten about AMD having fabs and how much of a shit show it was.
 
Retail data shows falling prices (and thus margins) for Intel desktop chips and assumedly same goes for server with their price cuts in the last year.
AMD has very likely overtaken or beginning to overtake Intel for margins.

Sales price isn't margin. Intel and AMD share CONSUMER CPUs at the lower and higher end now. The graph does not accounting for the vast majority of Intel's profits in SERVER CPUs which sell at a very high margin. AMD only holds about 17% of that market. This is why Intel still makes Billions more than AMD even though AMD has a good foothold on the consumer end.
 
You misunderstood, much of the technology that went into the 2000 and 3000 was a direct result of requests from the data centre and AI crowds, which unless they started doing completely custom cores for them much of those features and architectures were going to end up down in the gaming market. The biggest push for Ray Tracing wasn't coming from EA, or Epic, it's been from Disney, WB, and other animation/movie studios which then got Microsoft and NVidia involved which then brought it to games. So much of these things were done for work then they figured out how to use them for play.
The Nvida marketing push of RT on to gamers and gaming companies is a marketing coup. It borders on propaganda.
 
The Nvida marketing push of RT on to gamers and gaming companies is a marketing coup. It borders on propaganda.
Well they needed to create a market to up sell the features too somehow. But with upcoming changes in rendering techniques it’s overall a good thing.
 
Well they needed to create a market to up sell the features too somehow. But with upcoming changes in rendering techniques it’s overall a good thing.
Not saying it's bad, just misrepresented as actual ray tracing which is a joke and hyped over the top. But I digress. Off soap box.
 
My bad, I dont know why this popped up in my current news. delete
That article makes me feel worse about AMD's position, it means that their Nov launch was being prepped for as early as Jan of 2020 and this is what their efforts brought forth... Not exactly inspiring in that context.
 
Here it is, zen4 is 5nm , 5nm production to double.

https://www.tomshardware.com/news/tsmc-n5-capacity-expansion-plan

TSMC also planning SIX new Fabs in USA. Root Source is Taiwan News. Take it for what it's worth I guess.
That's sort of good that TSMC is doubling 5nm production taking it from its current 60,000 wafers a month to 120,000 wafers per month, but that means TSMC's entire 5nm capacity is currently less than AMD's allocation of 7nm (140K wafers per month), so once AMD gets their CPU's off 7nm and onto 5 I suppose that sort of increases supply assuming they don't then cut back their 7nm allocations accordingly.
 
Apple pays TSMC a lot of money to get access to those resources, nothing is stopping somebody else from offering TSMC more for their slot.

That's like saying "nothing is stopping you from trying to outbid Bill Gates on something he absolutely has to have"
 
Well there is only so many iPhones people want, if Apple always gets the lead node, so be it. As AMD moves to 5nm this Fall, Apple will eventually move to 4(2022)/3nm(2024). TSMC is going capacity building crazy, it will be a short time before supply meets demand,that's all.
 
Well there is only so many iPhones people want, if Apple always gets the lead node, so be it. As AMD moves to 5nm this Fall, Apple will eventually move to 4(2022)/3nm(2024). TSMC is going capacity building crazy, it will be a short time before supply meets demand,that's all.
That's the thing. People want iPhones and not need them. So many people buy the newest iPhone on release even of their last year's model does everything they need.
 
Last edited:
That's the thing. PeopLe want iPhones and not need them. So many people buy the newest iPhone on release even of their last year's model does everything they need.
Dance all you want about it but the point remains. There is a finite demand on everything.
 
They haven't reached it yet.
It was met a long time ago. Simply go to apple.com and order whatever you want. You act as if the iPhone encompasses 100% of TSMC capacity and need more. It's not the case and apple simply swallows the lead node. It's really not a big deal when simultaneously TSMC is expanding 5nm NOW. Put it together man.
 
That's like saying "nothing is stopping you from trying to outbid Bill Gates on something he absolutely has to have"
Exactly, AMD chose a node that was in extreme demand to release all their products, this is an issue they should have seen a mile away, and makes me very glad that NVidia chose Samsung, can you imagine if they were both on TSMC for this.... ick.
 
I thought TSMC *was* their first choice. They expected to go all-in when TSMC was like, "But ... AMD?" and sent Nvidia in a panic.

Ultimately I don't see AMD quickly spinning something up on Samsung for GPU production. Sure, they probably could do an ARM chip with RDNA and make a new market, but as long as the precursor materials are part of the bottleneck, they'd be competing against themselves in a way.
 
TSMC is AMD's ONLY CHOICE, as Apple an many others. It is expensive and time consuming to port to a different company's node. They probably underestimated demand BUT no one could have predicted a total output DROP because of shortages in base fab materials. - PERIOD. Obviously, TSMC made sure Apple got their chips FIRST, leaving scraps for the rest.
 
I get pings for the 6800 and 6900xt daily, so they're being produced.
But certainly noy in quantity for the retail market. They probably are going to miners also though not at the huge rate that Nvidia is selling them. You guys keep ignoring the real issue. When 85% of AMD's silicon wafers from TSMC are going to console chip production it simply does not leave enough silicon to make Epyc cpus, Ryzen cous, Ryzen APU's, and last on the priority list Radeon RX 6000 series gpu's. This was a highly prdictable situation. The contracts with SONY and Microsoft give them first claim to AMD;s 7nm silicon allocation from TSMC, even though the rate of profit on console chips is lower than AMD's cpu and gpu chips. AMD did not secure foundry space from Samsung on 5nm process which roughly equivalent or slightly better performing than 7nm TSMC. Yes Samsung is not TSMC as far as yields on large monolithic dies, as Nvidia is realizing now with their high end 3000 series. But console chips are not nearly as large as large as gou chips and yields would be much better. So AMD could have predicted this mess a year ago and worked with Samsung to produce their console chips but foolishly hoped against hope. Here we are screws as a result. Since AMD has zero loyalty to its enthusiast customer we should not have loyalty to them. Let us hope Intel continues to make great strides in improving their dedicated gpus so that more competition will inflict some pain on AMD and Nvidia and we might benefit with greater availability.
 
Back
Top