AMD: Moore's Law's end is near

[Hulk]

I remember reading that Intel has a lot of problems with their 22nm process. Hence the 16 months from Sandy to Ivy. Either they had a breakthrough that will help them get to 14nm or 14nm will be even harder.

The Intel of today is tighter-lipped than that of 6 years ago so I wonder what the Ivy Bridge defect scale looks like compared to this.
http://www.anandtech.com/show/2161/4

I have a feeling the game with silicon will be over around 5nm. I just don't see how they will get around quantum tunneling and the leakage it brings.

That being said there is still a lot they can do to make better performing parts.
While without the "Tick" there isn't additional die space for the "Tock" improvements they can still optimize various parts of the process even if the process as a whole doesn't shrink. And by the time we get to 5nm you're going to be able to fit a lot of transistors on a very small die.

I have a feeling Intel is racing to the smallest process as fast as possible because once they have the process down they have the bigger unknown solved. The smaller unknown I would think is the next architecture.

 

[mrmt]

We have the test now: what are the arguments Intel is going to use to sell this to Samsung or Apple?

Digital radio, with costs, die size and power scaling along with their nodes. Near threshold chips, that will make ARM chips look gas guzzlers. The most efficient manufacturing process the money can buy. Lots of top-notch architects that can design for you. Last, but not least, everything under the same roof.

Just don't count Intel out yet. They are in for a rough ride in the next 3-4 years, but if there is a company that can turn the tables on mobile it is Intel.

 

[VirtualLarry]

If anything, AMD gone might actually mean lower prices.

Is that what they teach in school in socialist countries like Denmark? That competition raises prices, and lack of competition lowers them?

Look at the US broadband internet market. I'm sure that theory worked there. *snicker*

 

[Hulk]

One question you have to ask yourself is why can Samsung, Qualcom and the rest compete or beat Intel in the low power market but not in the normal desktop space?

Intel has crushed every competitor that has come along over the years. Cyrix, NeXt, Motorola, and yes, sadly AMD. So why didn't (doesn't) Samsung or any of the others go after them in their arena? Too much of a head start? The money isn't there? Unfair business practices? I don't know. But I do know that except for the P4 Intel always had the best product.

Honestly I think the reason is because they can't compete with Intel. They pounced on a void in the market which is great for them and for the consumer. But they won't have it to themselves for long. And if history has showed us anything when it comes to what Intel can do to their competition they had better be prepared.

Right now in the phone and tablet market Intel isn't there. I mean the 800lb gorilla isn't there. They just sent a monkey in to have a look around.
But the gorilla is coming...

 

[R0H1T]

One question you have to ask yourself is why can Samsung, Qualcom and the rest compete or beat Intel in the low power market but not in the normal desktop space?

Well frankly because the desktop market as we know today isn't so much as a monopoly that is Intel rather a duopoly of "Wintel" from ~30yrs ago !

Intel has crushed every competitor that has come along over the years. Cyrix, NeXt, Motorola, and yes, sadly AMD. So why didn't (doesn't) Samsung or any of the others go after them in their arena? Too much of a head start? The money isn't there? Unfair business practices? I don't know. But I do know that except for the P4 Intel always had the best product.

Again because they have a history & legacy of 40+ yrs but if you think that the ARM/Android threat isn't real then you need to change your perspective.

Honestly I think the reason is because they can't compete with Intel. They pounced on a void in the market which is great for them and for the consumer. But they won't have it to themselves for long. And if history has showed us anything when it comes to what Intel can do to their competition they had better be prepared.

Why do you think Intel chose Android/Linux over Windows CE in the mobile market or Microsoft went with ARM for their mobile foray ?

Right now in the phone and tablet market Intel isn't there. I mean the 800lb gorilla isn't there. They just sent a monkey in to have a look around.
But the gorilla is coming...

This is a myth, "the bigger they are the harder they fall" is always a greater truth than Moore's law anytime & every time !

 

[MisterMac]

Easy there, just because you've been proven wrong doesn't mean you have to throw a tantrum !
For LG its 134 billion $ in 2011 ~

Samsung was US$ 247.5 billion in 2011 ~

So unless you have recent numbers just leave this argument ! As for the rest of it I've already said that your assumptions are not fit for a market that is contracting !

My bad - actually looked over and wrote a reply on LG's revenue (as you can see) to you earlier where you apparently didn't felt like responding.
(dunno how i ended up with the 211 number).

But your math still sucks you avoided ALL The other arguments i made.
It's not even 2x the size in revenue - so where did 2,5 to 3.0 come from?

Aka your baiting and almost fanboy'ish for AMD.


And that leaves us with the realization you had with mrmt now that - the impact will not be immense.

And that SOMEHOW - the whole marketshare AMD represents would go up in void smoke because AMD somehow doesn't exist anymore.

That is your analysis.

Do you know how retarded that sounds on a scale from 1 to 10?


PS:
x86 is not shrinking - server revenue is growing and so is shipments.
And that is what in a cloud world will keep Intel alive no matter what happens - albeit in handicapped state if they dont get into mobile.

 

[R0H1T]

My bad - actually looked over and wrote a reply on LG's revenue as you can see to you earlier where you apparently didn't wanna start up an argument you'd factually loose when comparing AMD to LG.

But your math still sucks you avoided ALL The other arguments i made.
It's not even 2x the size in revenue - so where did 2,5 to 3.0 come from?

My analogy wrt LG was just to show how LG as a giant in the electronics segment would be missed greatly, if they go bankrupt, likewise there will be an immense void in the x86+GPU market. I really can't find any other valid comparison here btw I was talking about 2013 & with Samsung being the largest mobile, dram, nand maker on the planet it isn't that much of a long shot ! If you want I'll try to find hard stats on that as well ?

Aka your baiting and almost fanboy'ish for AMD.


And that leaves us with the realization you had with mrmt now that - the impact will not be immense.

And that SOMEHOW - the whole marketshare AMD represents would go up in void smoke because AMD somehow doesn't exist anymore.

That is your analysis.

Do you know how retarded that sounds on a scale from 1 to 10?

Perhaps not in the grand scheme of things but with the x86+GPU market being a large subset of the semiconductor industry, the former will indeed be greatly impacted by their ouster !

PS:
x86 is not shrinking - server revenue is growing and so is shipments.
And that is what in a cloud world will keep Intel alive no matter what happens - albeit in handicapped state if they dont get into mobile.

The server market isn't growing as much as the desktop/laptop market is going down ! You can see from Intel's revenues that they had lower revenues YoY so let me break this one last time for you ~

1) Server market is growing.
2) Desktops/laptops are declining.
3) More mobile sales for Intel YoY.

So strictly speaking the traditional computing market is on the decline & even relatively, with increasing mobile revenues, the x86 market is not growing & even AMD says so !

I hope this will finally put to rest the argument that the x86 market isn't shrinking atleast for now !

 

[ShintaiDK]

Is that what they teach in school in socialist countries like Denmark? That competition raises prices, and lack of competition lowers them?

Look at the US broadband internet market. I'm sure that theory worked there. *snicker*

You cant compare it to broadband or other utility style services. First flaw in your lack of economic understanding.

Second flaw, higher volume means lower development costs. And price needs to be in a range that offers maximum volume/profit ratio. That would be current prices.

Third flaw, Intel needs to give you a reason to buy a CPU, not the other way around. Else you simply keep the previous CPU.

And dont talk about socialism when you dont have a clue what it is. And seems you skipped capitalism as well in school.

 

[Homeles]

One question you have to ask yourself is why can Samsung, Qualcom and the rest compete or beat Intel in the low power market but not in the normal desktop space?

Because they don't have an architecture suitable for desktop use. Likewise, Intel does not have an architecture suitable for smartphones and tablets, so Intel's being beaten. They were able to shrink down their aging Atom into the thermal range of smartphones and tablets, but it's not truly "meant" to be there. Neither is Ivy Bridge. Haswell and Silvermont will be the first ones that are really "meant" for that market.

As far as I know, ARM has no intention of making performance CPUs. They want to stick with the mobile side. On the other hand, Intel wants to have their hands in both cookie jars at the same time.

 

[Idontcare]

Well, Intel is aiming for 5 nm or so in less than a decade. That strikes me as incredibly aggressive, and probably not realistic. Not only will process technology innovations be needed, the geometry of the transistors is going to have to keep changing, because I doubt we can scale gate thicknesses down that much.

The process node label is just that, a label. It is not intended to be a numerical reference to anything with physical dimensionality.

We already do this with gate oxides, have for more than 15yrs. We call it "EOT" for effective xide thickness. We measure the electrical characteristics of the oxide and then assign a fictious value to it for its thickness in terms of how thin the oxide would be if it were made from traditional silicon dioxide.

So you get silly numbers like "3Å EOT" which defies physics as silicon dioxide cannot physically be any thinner than 7Å, and so on.

The same is true of node labels. There is nothing in Intel's "22nm node" that is 22nm in the sense of what the node label conveys to the lay person. Instead what they are capturing and communicating with the node label is the fact that the electrical performance has significantly improved over that of the prior node (labeled 32nm in this case).

So what will the "5nm node" look like physically? Nothing like an existing node scaled down to 5nm dimensionality. But electrically it will perform as if one had scaled the physical dimensions as such. And that is all they mean to communicate with the node label.

The scaling of electrical characteristics, aka "parametrics", is what will continue unabaited. But physical scaling in the traditional sense is coming to end, has been for the better part of a decade now.

 

[BallaTheFeared]

Wait wut, come again...

 

[Hulk]

Because they don't have an architecture suitable for desktop use. Likewise, Intel does not have an architecture suitable for smartphones and tablets, so Intel's being beaten. They were able to shrink down their aging Atom into the thermal range of smartphones and tablets, but it's not truly "meant" to be there. Neither is Ivy Bridge. Haswell and Silvermont will be the first ones that are really "meant" for that market.

As far as I know, ARM has no intention of making performance CPUs. They want to stick with the mobile side. On the other hand, Intel wants to have their hands in both cookie jars at the same time.

The answer to my question is that in my opinion is that they knew that the full brunt of Intel wouldn't come down on them while Intel was occupied in the desktop/mobile space. What were scraps that Intel was content to let them have 6 years ago is now where the market is moving. I don't think they will have any more success in the long term competing against Intel in that market than they would have in the desktop/mobile market. Intel is coming for them and as much as I want a long drawn-out competitive battle I fear the ARM guys will be out thought (better Intel micro-architecture), out fabbed (much much better process tech), and out marketed in less than 5 years.

 

[Homeles]

That's how I see it as well. Maybe not quite as quickly as 5 years, but I see Intel winning that segment without much of a fight, except from Qualcomm and Apple.

Wait wut, come again...

Let's replace node names with Intel's code names.

45nm = Nehalem
40nm = Westmere
32nm = Sandy Bridge
28nm = Ivy Bridge
20nm = Haswell

Full node shrinks are tocks. Half node shrinks are ticks. 45nm doesn't mean anything -- it's a code name, just like Nehalem. Westmere is better than Nehalem in the same sense that 40nm is better than 45nm, but you don't know how much better.

All 40nm means is that it's better than its predecessor, 45nm. It really give tell you how much better density will be, nor does it tell you what the transistor gate length is.

 

[R0H1T]

The answer to my question is that in my opinion is that they knew that the full brunt of Intel wouldn't come down on them while Intel was occupied in the desktop/mobile space. What were scraps that Intel was content to let them have 6 years ago is now where the market is moving. I don't think they will have any more success in the long term competing against Intel in that market than they would have in the desktop/mobile market. Intel is coming for them and as much as I want a long drawn-out competitive battle I fear the ARM guys will be out thought (better Intel micro-architecture), out fabbed (much much better process tech), and out marketed in less than 5 years.

That's like way over the top ! You think that Samsung(~6x group revenue) or Qualcomm(fastest growing chipmaker) even the likes of Nvidia/TSMC are just a bunch pushovers ?

 

[Fjodor2001]

If we'll hit a wall in 10 years using silicon they better start researching other options. Moving to some completely new process technology is not exactly like a standard node shrink that can be done in 2 years or so. It seems like there should almost be working prototype chips by now if we're to smoothly transition into a post-silicion era in 10 years.

I also wonder if this means it's open for completely new companies to enter the process tech arena? Who knows, the next gen process technology could perhaps just as well be invented in a research lab/facility at some university. Then it's of course something else to take that into full scale production. But if you have a working concept and patents it should be possible to get venture capital to do that. If there is to be such a breaking point, it's not unlikely to happen when transitioning to a completely new process tech. The reason is that then everything is new and past experiences and established practices may not matter as much as when simply improving upon existing technology which is the case with a normal node shrink.

 

[NTMBK]

One question you have to ask yourself is why can Samsung, Qualcom and the rest compete or beat Intel in the low power market but not in the normal desktop space?

Intel has crushed every competitor that has come along over the years. Cyrix, NeXt, Motorola, and yes, sadly AMD. So why didn't (doesn't) Samsung or any of the others go after them in their arena? Too much of a head start? The money isn't there? Unfair business practices? I don't know. But I do know that except for the P4 Intel always had the best product.

Honestly I think the reason is because they can't compete with Intel. They pounced on a void in the market which is great for them and for the consumer. But they won't have it to themselves for long. And if history has showed us anything when it comes to what Intel can do to their competition they had better be prepared.

Right now in the phone and tablet market Intel isn't there. I mean the 800lb gorilla isn't there. They just sent a monkey in to have a look around.
But the gorilla is coming...

The desktop market is based on entrenched software. People want compatibility above all else. Samsung and Qualcomm don't have x86 licenses, so they can't access that market. People have tried before with Linux based netbooks and nettops, and they failed.

 

[Homeles]

That's like way over the top ! You think that Samsung(~6x group revenue) or Qualcomm(fastest growing chipmaker) even the likes of Nvidia/TSMC are just a bunch pushovers ?

Relative to Intel, yes.

 

[Hulk]

That's like way over the top ! You think that Samsung(~6x group revenue) or Qualcomm(fastest growing chipmaker) even the likes of Nvidia/TSMC are just a bunch pushovers ?

I don't think they are pushovers at all. But based on the 40+ year history of Intel and the past 20 years of which I've been a keen observer that is my opinion. The manufacturers you mentioned are filling a void. A gas expands quickly in a void.

You've got to ask yourself why no one, no one on the freaking planet has been able to build a better x86 processor than Intel? And we know there is money to be made with x86 processors. You can claim they started first, or they had/have a monopoly with MS, or whatever, but in the end they still have far and away the best product. Only AMD is even close.

With all of their revenue Samsung has never even dipped a toe into Intel's water. In addition, the rate of increase of Intel's GPU's has been astounding over the last few generations. Another case of expanding into a vacuum. And by that I mean from where they came GPU-wise, which was crap.

There's nothing like a heavyweight fight and this is going to be a good one. I just don't see anyone having the design talent/experience, fabs, and marketing savvy to compete with Intel in any microprocessor market they decide they want. What should be even scarier for the likes of their competition you mentioned is the fact that they don't seem to be developing their own ARM parts. Either they are really stupid and are about to make a possibly fatal miscalculation or they are really smart.

 

[Charles Kozierok]

The process node label is just that, a label. It is not intended to be a numerical reference to anything with physical dimensionality.

Interesting info, IDC, thanks.

Do you know when they started playing these sorts of games? I know that at one point the number did refer to an actual dimension (gate length, I believe.)

 

[Idontcare]

Interesting info, IDC, thanks.

Do you know when they started playing these sorts of games? I know that at one point the number did refer to an actual dimension (gate length, I believe.)

It meant something physical up to the 0.35um node.

In the far future, decades from now, historians will look back on this era and label the 0.35um node as being the timeframe at which true traditional scaling broke down.

What we've been living in since then is a world of electrical scaling. As consumers we readily accept this reality when it comes to our MLC and TLC based NAND-flash SSDs and "6GHz" QDDR5, but we aren't conditioned to recognize it in our CPUs (yet).

But hindsight will see that the history books are written correctly. I'm not worried about the record being set straight any time soon

 

[TuxDave]

Interesting info, IDC, thanks.

Do you know when they started playing these sorts of games? I know that at one point the number did refer to an actual dimension (gate length, I believe.)

From my experience, the last time I saw something on the CAD screen match something related to the process was on dual oxide 90nm from ST Micro. In that case, the minimum drawn gate length measured out to be 90nm on the CAD screen. The post-transformed mask is probably warped in some voodoo fashion.

Then I jumped straight to 45nm and never saw any matches between CAD and process.

 

[Thegonagle]

Bah! They'll just add more cores.

More cores for everyone! :thumbsup:

 

[Exophase]

Meanwhile, Intel says:

Mark Bohr who overseas much of Intel’s processor research, was quoted as saying, “The end of Moore’s Law is always 10 years away. And, yes, it’s still 10 years away.”

()

This actually makes sense, but not in the way most will probably interpret it..

You don't tend to reach hard limits abruptly. Usually the mechanism that limits it will also make it harder and harder to approach that limit. So shrinking may indeed never end, but approach an asymptotic line more and more slowly until it no longer makes economic sense for anyone or something else displaces it.

But that doesn't mean you can't identify that there are hard limits. At least in the universe we live in.

 

[VirtualLarry]

And seems you skipped capitalism as well in school.

So let me get this straight, someone from a socialist country, tells me that less competition will lower prices, and *I* am the one that is missing an education on capitalism?

LOL!

Edit: So basically, your argument is this - AMD, by competing with Intel, has conspired to keeping Intel prices higher than they could be, because it prevented Intel from getting the volume sales that would come from owning 100% of the market, which would lower Intel's R&D costs per unit sold, thus Intel would have lowered their prices? Am I correct? Doesn't anyone else with an education in capitalism, see the flaws in this picture?

 

[Charles Kozierok]

90 nm or so seems to be about what I've found in my research as well. 350 nm takes us back, what, 20 years? I'd be surprised if they started tinkering with the node definitions that early, but I'm not a process engineer.