Intel’s Stacked Nanosheet Transistors Could Be the Next Step in Moore’s Law

[TOMPPIX]

Wasn't intel also doing the little cores and big cores, is so couldn't they just put the slow cores under the fast cores?

[straight_stewie]

10 hours ago, Kisai said:

And it doesn't matter if AMD or Intel are doing good or poorly in this regard, Intel is behind, and because their fingers are in a lot more diverse things (eg storage, networking, FPGA's, etc) they can actually afford to weather the cluckstertruck that their 10nm process is. But at this point if they're not putting desktop cpu's on 10nm, it doesn't matter, the marketing has spoken for Intel.

You're not understanding what the phrase "conducting good business" means, but that's ok.

For whatever reason, people on this forum only seem to care about one single product line that Intel offers, and can't seem to see the bigger picture going on here.

Quote

AMD's chip shortages proves just how far Intel is behind when AMD is charging more for a chip in the same performance tier and has been entirely unavailable the entire year.

You're proving my point that AMD is incapable of performing over the long run. They are entirely dependent on another business, in which they are not even a controlling partner, for all of their manufacturing capabilities. Hence, they can't keep up with demand for their product, hence they can't really grow.

Beyond that, AMD is now stuck at TSMC 7nm (which, for reference, is competitive to Intel's 10nm) for the foreseeable future because TSMC has decided to sell basically all of their 5nm capabilities to Apple, essentially forever. That opens up all kinds of opportunities for Intel, like moving straight to 5nm, even if it means spinning their wheels for another couple of cycles, and AMD simply won't be able to do anything about it.

AMD simply does not have the cash reserves necessary to fix their supply line issues. Beyond that, they won't, for a long time, even if they magically stay profitable for longer than is likely. That entire decade of operating at a loss forced them to take on a lot of debt.

 

The debt-to-equity ratio is calculated by taking the total claims against a businesses assets by creditors and dividing it by the total claims against a businesses assets by it's investors (owners). Basically, a debt-to-equity ratio tells us how much of each dollar of the value of a business each business owner gets after fully selling all assets at book and settling all debts. Lower is better, and a debt-to-equity ratio of 1.0 or greater means that there is not enough value to pay off all debts.

Now the good stuff. AMD's debt-to-equity ratio is 1.68, meaning that the business is worth -$0.68 dollars for every dollar invested. In contrast, Intel's debt-to-equity ratio is 0.52, which means that the business is worth $0.48 cents for every dollar invested. Of course, the businesses are worth more than that, but the debt-to-equity analysis only accounts for the case where you immediately fold a business, sell all of it's assets at book value, and distribute the proceeds.

Basically, what this tells us, is that AMD's investors own negative 68% of their business, while Intel investors own positive 52% of their business.

Now, I don't claim to be the best investor in the world, but I will put my money where my mouth is on that one. AMD is in a hole. One that will take them at least the next 3 years to crawl out of if they can maintain their current performance. In the meantime, Intel is cash flush and moving forward, just in ways that the PC enthusiast group is having trouble seeing right now. 

[leadeater]

8 hours ago, straight_stewie said:

They are entirely dependent on another business, in which they are not even a controlling partner, for all of their manufacturing capabilities.

So? Same is true for Apple, didn't prevent them from being successful.

 

8 hours ago, straight_stewie said:

because TSMC has decided to sell basically all of their 5nm capabilities to Apple, essentially forever.

Yea that is just a rumor, I highly doubt it's actually what has been talked about and AMD wouldn't have moved to 5nm any quicker regardless anyway. Also not essentially forever, TSMC would never make such an agreement, everrrrr. They have contract for current capacity for a certain length of time, TSMC is increasing 5nm capacity as was planned like every other node transition. You do know the same situation happened with 7nm right? Apple was first customer and did nothing to block TSMC being able to supply other clients. Right now AMD is the largest company TSMC works with in TSMC production capacity allocation.

 

8 hours ago, straight_stewie said:

AMD simply does not have the cash reserves necessary to fix their supply line issues. Beyond that, they won't, for a long time, even if they magically stay profitable for longer than is likely. That entire decade of operating at a loss forced them to take on a lot of debt.

AMD for many years now have been actively paying down debt as fast as they can, however a company of the size and scale of AMD having debt is normal. End of 2015 AMD had $2 billion in long term debt liability, right now less than $400 million. 2018 to 2019 alone they paid down more than $600 million of that long term debt, they are doing just fine.

 

So no they are not taking on debt, entirely the opposite.

 

On the other hand Intel's total debt is $36 billion, up $11 billion from last year. So which of the two is actually taking on debt? Certainly not going to say Intel is danger of anything simply because they have debt or have increased it.

[Quackers101]

When will the sheets hit the streets?

so we can see how they roll.

but cool indeed.

[CarlBar]

1 hour ago, leadeater said:

So? Same is true for Apple, didn't prevent them from being successful.

 

Yea that is a just a rumor, I highly doubt it's actually what has been talked about and AMD wouldn't have moved to 5nm any quicker regardless anyway. Also not essentially forever, TSMC would never make such an agreement, everrrrr. They have contract for current capacity for a certain length of time, TSMC is increasing 5nm capacity as was planned like every other node transition. You do know the same situation happened with 7nm right? Apple was first customer and did nothing to block TSMC being able to supply other clients. Right now AMD is the largest company TSMC works with in TSMC production capacity allocation.

 

AMD for many years now have been actively paying down debt as fast as they can, however a company of the size and scale of AMD having debt is normal. End of 2015 AMD had $2 billion in long term debt liability, right now less than $400 million. 2018 to 2019 alone they paid down more than $600 million of that long term debt, they are doing just fine.

 

So no they are not taking on debt, entirely the opposite.

 

On the other hand Intel's total debt is $36 billion, up $11 billion from last year. So which of the two is actually taking on debt? Certainly not going to say Intel is danger of anything simply because they have debt or have increased it.

 

Thanks for addressing all of that, i wanted to take a swing at it, but was having trouble finding a good starting point. Plus there's some stuff in there i didn't know.

 

I'd also add that TSMC doesn't want to be dependent on apple for their business, it's not in their interests to let apple cutoff AMD and NIVIDIA. And from what we know late 2021 is the earliest we'll see Zen 4. Whilst apple has a lot of stuff lined up for the early and mid part of the year. So Apple's buying up of capacity may not actually affect AMD all that much, they may even have had a prior contract for that other 20% of 2021's capacity in place before Apple ever came along.

 

Also not having such a diverse business means AMD has much lower overheads, so having smaller absolute revenues isn't necessarily a killer for them. They can survive on a lot less income.

 

Thats not to say Intel is in rouble. Even with all the capacity of every other fab capable of manufacturing a high performance CPU or GPU being used to produce competing products Intel will still have the majority market share by a decent margin because those other nodes combined can't produce enough chips to supply enough of the market to take it away from Intel. it can make gains, it can make Intel's life hard, but it can't kill it.

[Spindel]

As I said in some other thread a while ago this is the future for higher transistor density BUT...

 

...this will have issues with heat since you will have multiple layers of heat producing components, while the top one will get full advantage of the cooler the lower layer/layers will not. 

 

In the thread I mentioned this earlier I was jokingly saying that the solution would be silicon carbide since it can operate at several hundred degrees (Celsius) and thus you could allow the internals of the CPU to become really hot. 

[Quackers101]

just get water cooling on the small scale

[Kisai]

2 hours ago, Spindel said:

As I said in some other thread a while ago this is the future for higher transistor density BUT...

 

...this will have issues with heat since you will have multiple layers of heat producing components, while the top one will get full advantage of the cooler the lower layer/layers will not. 

 

In the thread I mentioned this earlier I was jokingly saying that the solution would be silicon carbide since it can operate at several hundred degrees (Celsius) and thus you could allow the internals of the CPU to become really hot. 

Honestly, why not just go back to the Slot 1 solution, and put heat-sinks on both sides of the CPU. For those who aren't aware of how the Slot 1 CPU's were setup, the cpu was in the center of one side of the "cpu" module, and the L2 cache was at half speed and on external chips inside the package. So perhaps there's a way to put the CPU die itself have a surface on both sides, while having the pins along the edge of this "slot" type cpu card. The current ATX motherboard designs don't really permit putting anything on the backside of the motherboard which is why you'd consider this. Even then to double-side the cpu would still require some way of attaching to the motherboard that would make you ask, why not just put two CPU dies in the module in the same kind of MCM arrangement, only on directly opposite sides so the infinity fabric doesn't have to go off the cpu module.

 

I know there's a lot of mechanical reasons why this is unreasonable, but even looking at threadripper, it seems like there's not a lot of options for stacking the die without taking more and more space on the motherboard. May as well just have the CPU be the motherboard, and design the cooling around it.

[Spindel]

1 hour ago, Kisai said:

Honestly, why not just go back to the Slot 1 solution, and put heat-sinks on both sides of the CPU. For those who aren't aware of how the Slot 1 CPU's were setup, the cpu was in the center of one side of the "cpu" module, and the L2 cache was at half speed and on external chips inside the package. So perhaps there's a way to put the CPU die itself have a surface on both sides, while having the pins along the edge of this "slot" type cpu card. The current ATX motherboard designs don't really permit putting anything on the backside of the motherboard which is why you'd consider this. Even then to double-side the cpu would still require some way of attaching to the motherboard that would make you ask, why not just put two CPU dies in the module in the same kind of MCM arrangement, only on directly opposite sides so the infinity fabric doesn't have to go off the cpu module.

 

I know there's a lot of mechanical reasons why this is unreasonable, but even looking at threadripper, it seems like there's not a lot of options for stacking the die without taking more and more space on the motherboard. May as well just have the CPU be the motherboard, and design the cooling around it.

Looking at the number of pins on (at least) modern x86 CPUs I have a hard time seeing how doublesided cooling would be achieved. At least back in the days with Socket A and similar I could imagen it (like 4 rows of pins along the edges) but modern FPGA or LPGA not so much. 

[leadeater]

4 hours ago, Spindel said:

...this will have issues with heat since you will have multiple layers of heat producing components, while the top one will get full advantage of the cooler the lower layer/layers will not. 

I doubt that, this isn't like die stacking it's just introducing height to the transistors at an incredibly small scale. Looking at the graphic provided by Intel the actual height doesn't appear to much different at all though. Instead of the gates of the transistors being side by side they are top and bottom and it's likely that the transistors themselves using the nanoribbons may actually be better at heat dissipation, if only very slightly. Doubling the density might be a bigger issue for heat, heat density is already a problem now. Improving the PMOS might help with this too as it's less efficient than NMOS.

 

If the nanoribbons and PMOS improvements allow lower operating voltages that'll also help, not sure on that one though. That one depends on how hard you want to push the clocks so any improvements might be counteracted by pushing extra performance.

 

Intel 14nm products are only running hot now because of how hard they are pushing that node to stay competitive, earlier generation 14nm products were actually cooler than the previous 22nm products.

[straight_stewie]

On 1/6/2021 at 6:43 PM, leadeater said:

however a company of the size and scale of AMD having debt is normal. End of 2015 AMD had $2 billion in long term debt liability, right now less than $400 million.

Of course having debt is just one of the costs of doing business.

What you missed was "debt-to-equity". If you have more debt than you do equity, you're in a pickle. By my math, it will take AMD three years to finish paying off their debts, and that's assuming that things keep looking up for them. Numerous things could happen in the meantime (like Alder Lake or ARM. Keep in mind that Alder Lake will be real competition even though AMD is on 7 and Intel will be on 10, because Intel's 10 gives them a transistor density higher than TSMCs 7...).

Think about it like this: You are being offered the opportunity to buy a portion of a company. When you buy a portion of the company, you immediately get some amount of the companies value, and that value is not really related to the cost of acquiring it. Would you rather spend money on the company that gives you a dollar and fifty cents for every dollar you spend to by a portion of the company, or spend money on the company that costs you a dollar and sixty cents for every dollar you spend to buy a portion of the company?

What I'm saying is that while the value of AMD's common stock may go up and down based on, well, nothing really, at the end of the day it's really only worth whatever percentage of the book value of the company is equivalent to whatever percentage of outstanding common stock you own, minus that portion of equity owed to debtors, minus that portion of the equity owed to preferred shareholders. And right now, AMD has negative value even though it looks like they are performing well right now.

Intel has room to acquire more debt. AMD does not. AMD has a ways to go to become a worthwhile company to be interested in, Intel is more profitable now than they ever have been. Even with Intel taking on the additional debt you mentioned, their debt-to-equity ratio has still decreased, at a rate comparable to how fast AMD is decreasing theirs.

Like I've been saying all along, the socketed desktop processor is not all there is to this analysis, and that market will likely not make or break Intel over the next three to five years (hence why they aren't really pushing it right now). With the end of Moore's Law at hand, and competition from other architectures creeping in, anything could happen here. But, at the moment, Intel is still ahead of AMD overall.

[leadeater]

1 hour ago, straight_stewie said:

By my math, it will take AMD three years to finish paying off their debts, and that's assuming that things keep looking up for them.

Finish, no company of these sizes operates at zero debt. If AMD can pay off 600 million in a single year shifting less product volume than they are now by a lot I'm sure they could pay off much more than that IF they wanted to.

 

1 hour ago, straight_stewie said:

What you missed was "debt-to-equity". If you have more debt than you do equity, you're in a pickle

No I didn't, it's fairly irrelevant for a company like AMD. The IP alone that AMD controls is worth many times more than their debt so moot diversion.

 

1 hour ago, straight_stewie said:

Even with Intel taking on the additional debt you mentioned, their debt-to-equity ratio has still decreased

Not correct, not for that year, the year they took on 11 billion in debt it increased, but like I said this is such a useless thing to be looking at for AMD and Intel, and Nvidia, and Apple etc etc.

 

Edit:

Also AMD's total assets is 7 billion and total liabilities 3 billion so you're incorrect about that anyway.

Intel's just so it's here too, 145 billion vs 70 billion

[straight_stewie]

18 minutes ago, leadeater said:

No I didn't, it's fairly irrelevant for a company like AMD. The IP alone that AMD controls is worth many times more than their debt so moot diversion.

 

You do realize that AMD can't really sell their x86 IP, right?

The only IP they could sell would be x64 related, and the only really interested buyer would be Intel themselves...

AMD is only still allowed to make x86 processors because Intel looks the other way so that AMD will return the favor and allow Intel to make x64 processors. Had Itanium taken off, Intel would have challenged AMD on the x86 front right then and there.

If AMD tries to liquidate their x86 IP, Intel is sure to challenge, hard.

The AMD-Chinese joint venture only works because the Chinese partners are only allowed to build the processors, not design their own with the IP. It's just that, unlike TSMC, the Chinese partners make money on the back end, not the front end (where TSMC profits from  manufacturing the processors, the Chinese companies profit from directly selling the processors).

I've excluded the VIA complication for obvious reasons. VIA has not been, and likely never will be, a worthwhile competitor in the industry.

[leadeater]

22 minutes ago, straight_stewie said:

You do realize that AMD can't really sell their x86 IP, right?

Yes they can, under your hypothetical situation of them doing under it'll be acquired by anyone that wishes to acquire AMD or any parts of it under such deals that may take place. Also AMD owns a heck of a lot of very valuable IP other than x86, ever notice the lack of GPU competitors too?

 

22 minutes ago, straight_stewie said:

and the only really interested buyer would be Intel themselves...

Very much doubt that.

 

22 minutes ago, straight_stewie said:

AMD is only still allowed to make x86 processors because Intel looks the other way so that AMD will return the favor and allow Intel to make x64 processors.

Rubbish, they both have active cross license agreements, none of this "looking the other way". Intel would immediately die if AMD retracted that agreement, so would AMD for that matter. Neither have the greater hand of cards here.

 

But AMD is very much not in bad shape as a business as you've been trying to make out, FAR from it. You'd have a point if it were 2006-2009 but we're not living in 2006-2009 anymore.

 

But you're certainly not correct about the ratios you've mentions anyway, but this is my last post about this. You're welcome to go look at the data again, but I won't be discussing it more here as it's off topic.

https://www.macrotrends.net/stocks/charts/AMD/amd/financial-ratios

https://www.macrotrends.net/stocks/charts/INTC/intel/financial-ratios

Edited January 8, 2021 by leadeater

[straight_stewie]

1 minute ago, leadeater said:

But AMD is very much not in bad shape as a business as you've been trying to make out, FAR from it. You'd have a point if it were 2006-2009 but we're not living in 2006-2009 anymore.

AMD operated at a loss all the way up to 2018... So it's really 2006-2018. But whatever. Bet your money on AMD, what do I care?

[Brooksie359]

On 1/3/2021 at 6:42 AM, J-from-Nucleon said:

I dunno about rocket lake tho, I kinda feel like Rocket lake is essentially a stopgap for Intel and they're putting more of their efforts into Alder Lake. They litterally announced it via a Blogpost (If my memory serves me right)......A BLOGPOST! But anyway, I'll be still be looking at it hopeful

Intel has been releasing slightly improved versions of skylake for 5 generations at this point and rocketlake is the first time they have actually brought a new architecture to the desktop market. I imagine part of the problem is they had designs lined up for 10nm but when 10nm was delayed over and over again they had to release skylake based architectures over and over again. I am actually looking forward to Rocket lake as it seems like it will have the first real improvement in ipc for Intel in a long time. That added with their huge fab capacity that allows them to have plenty of supply to meet demand i don't think they will be hurting anytime soon.