How long can intel maintain such a large process lead over rivals?

[AtenRa]

Lets summon up. Glofo ships 32nm. Samsung ships 32nm, TSMC ships 28nm, Intel ships 22nm.

When did the above companies ship HKMG? Trigates?

Intel is around 3½ years in front of TSMC if you count all technologies. Samsung and Glofo is just hopeless behind with 4+ years behind.

Intel's first 32nm HKMG CPU was released on January 2010 (Westmere Core i3 5xx).
AMD's first 32nm HKMG SOI CPU (GloFo) was released on June 2011 (Llano)

That is less than 2 years

 

[pelov]

x86 may not be clean or elegant, but it gets the job done.

For how much longer?

Intel still hasn't learned a very important lesson and it's one that the ARM consortium knows all too well: people care about price and user experience above all else. Both Intel and Microsoft charge far too much for their products and licenses and as a result ARM and Linux have taken over the low end of mobile. Though Intel has entered the fray, the profit margins on Intel's chips are much higher than what others charge and on the low end (phones/tablets) it doesn't reason to use their chips and designs because it means the OEMs make less money. Microsoft has to tackle with the same issue because they charge for licensing whereas Linux-based alternatives -- and even iOS though not Linux -- are essentially free. What you end up with is a product that costs much more than a competing ARM/*nix alternative and the user really can't tell the difference between the two. Tizen (Intel/Samsung's baby OS) is a direct result of Microsoft tax just as ARM/MIPS growing into big boy trousers will be a direct result of Intel x86 tax.

Unless Intel starts lowering their margins or designing products from start-to-finish like Apple does (fabs included), Intel will soon get hit with a dose of reality: x86 isn't all that special

 

[2is]

People have been predicting x86 demise back when macs were stilling using PowerPC processors. X86 isn't going anywhere.

 

[Kenmitch]

I'm guessing equipment costs would take a huge chunk of the pie.

I would think the equipment costs are huge!

I watched a special on TV last night about the Lego empire. There was one plastic injection mold that cost 350K to produce which in turn only makes one of the items in the set.

 

[Charles Kozierok]

For how much longer?

That's the $64 billion question.

And exactly my point made earlier: Intel dominates in the x86 segment but that doesn't mean it is an all-conquering behemoth. It still has plenty of challengers, and markets are arguably evolving away from its core strengths.

 

[Arzachel]

For how much longer?

You're severely underestimating what marketing can do. Doesn't matter how competitive a product is, if the consumer thinks it's better.

Or an army of lawyers for that matter. Have you seen how MIAA/RIAA are struggling to stop the advance of tech? Apple vs Samsung? I'm pretty sure that x86 will be here long after it'll have outlived their stay. (It already has)

 

[ShintaiDK]

Intel's first 32nm HKMG CPU was released on January 2010 (Westmere Core i3 5xx).
AMD's first 32nm HKMG SOI CPU (GloFo) was released on June 2011 (Llano)

That is less than 2 years

Intel started shipping hkmg at 45nm, not 32nm. Intel used HKMG with Penryn aka 45nm Core 2. Or 2007. Thats 4 years.

 

[ShintaiDK]

For how much longer?

Intel still hasn't learned a very important lesson and it's one that the ARM consortium knows all too well: people care about price and user experience above all else. Both Intel and Microsoft charge far too much for their products and licenses and as a result ARM and Linux have taken over the low end of mobile. Though Intel has entered the fray, the profit margins on Intel's chips are much higher than what others charge and on the low end (phones/tablets) it doesn't reason to use their chips and designs because it means the OEMs make less money. Microsoft has to tackle with the same issue because they charge for licensing whereas Linux-based alternatives -- and even iOS though not Linux -- are essentially free. What you end up with is a product that costs much more than a competing ARM/*nix alternative and the user really can't tell the difference between the two. Tizen (Intel/Samsung's baby OS) is a direct result of Microsoft tax just as ARM/MIPS growing into big boy trousers will be a direct result of Intel x86 tax.

Unless Intel starts lowering their margins or designing products from start-to-finish like Apple does (fabs included), Intel will soon get hit with a dose of reality: x86 isn't all that special

Problem is ARM cant compete with x86. x86 can compete with ARM. We already see Medfield smartphones shipping in large volume. ARM already lost. Same with Windows 8 for ARM. Almost all tablet makers went for the x86. Because Windows on ARM is useless. Then you could just use another OS. AMD also knows this, thats why they are loving the sales of bobcat CPUs.

 

[KompuKare]

People have been predicting x86 demise back when macs were stilling using PowerPC processors. X86 isn't going anywhere.

Maybe. But the need to be backward compatible with a 30-40 year old design hurts. It's like so-called junk DNA really. Some may be useful but most dead-end sequence left over from evolution. But then evolution is 'dumb'. Let me clarify that bit. Obviously anything which gets in the way of fitness can be selected against but by dumb I mean more the localised 'valleys and peaks' problem: while design-wise a better design may be possible to get there the 'unfit' valley would have to traversed and that cannot usually happen because in the meantime the individual would be less fit. Catch 22.

The point is, with computing we are not reliant on random mutations and selection but we (humans) can use 'intelligent design'.

Problem is a lot technology (and especially x86) has tons of legacy baggage on-board and if someone was rationally going a clean design now that stuff would not be included. It's only there due market inertia.

But imagine if 100 or more years from now, computers still had hardware to rename registers, deal with 64KB segments, real mode etc. Truly crazy. And that doesn't just apply to CPUs but also OSs, C-syntax programming languages etc.

 

[ShintaiDK]

Personally I dislike x86 more than the average, including its terrible bastard spinoff x64. But lets face it. x86 aint going anywhere, its expanding. In a perfect world we would all use IA64 or something now. But it aint, its capitalism and its business. Its just a matter of time before x86 is essentially world dominating.

The same legacy hurle its also its strength.

We dont live in a world that can afford technical pure designs just for the sake of it.

 

[pelov]

You're severely underestimating what marketing can do. Doesn't matter how competitive a product is, if the consumer thinks it's better.

Isn't that how Apple works? They've been overpricing underwhelming hardware for ages and it's done them wonders. But the difference is you can't market x86 in the same way you can glossy white plastic. x86 isn't sexy and doesn't wear black-brimmed glasses.

Unless Intel starts producing products from theory to implementation all on their own, Apple style + fabs, the OEMs won't play their games for much longer. When you consider Win8 + Intel tablets will cost at least $100 higher than a competing ARM design due to licensing alone (not including other hardware), that means OEMs are losing $100 in profits. There's also a far cheaper alternative to both of those that is just as viable and allows OEMs to dictate their own pricing instead of carrying those two giants along as leeches. I know I'd rather have a far larger SSD or better high-res screen on an ARM tablet than win8+x86 and get stuck with crappier hardware.

Problem is ARM cant compete with x86. x86 can compete with ARM. We already see Medfield smartphones shipping in large volume. ARM already lost. Same with Windows 8 for ARM. Almost all tablet makers went for the x86. Because Windows on ARM is useless. Then you could just use another OS. AMD also knows this, thats why they are loving the sales of bobcat CPUs.

I think the 64-bit hurdle is the biggest one ARM has to cross. They've already patched some holes with Thumb and VFP but they've definitely got a long way to go. The biggest advantage that it carries is the low price and that's something everyone can get behind. Qualcomm is a perfect example of a company that takes a tiny little RISC ISA and does some incredible things with it. At this point, Intel is far more worried about ARM advancements into desktop/server than they are about AMD's new architectures and GPUs.

 

[AtenRa]

Intel started shipping hkmg at 45nm, not 32nm. Intel used HKMG with Penryn aka 45nm Core 2. Or 2007. Thats 4 years.

Heh, i could say that Intel is 10 years behind because AMD started shipping SOI with Athlon 64. Intel still hasn't, up to this day

By your logic, AMD (GloFo) has the lead now because it has both HKMG and SOI. :twisted:

 

[Charles Kozierok]

Heh, i could say that Intel is 10 years behind because AMD started shipping SOI with Athlon 64. Intel still hasn't, up to this day

By your logic, AMD (GloFo) has the lead now because it has both HKMG and SOI. :twisted:

Intel made a deliberate decision not to use SOI. It's not a straight advance -- it has both pros and cons.

 

[ShintaiDK]

Heh, i could say that Intel is 10 years behind because AMD started shipping SOI with Athlon 64. Intel still hasn't, up to this day

By your logic, AMD (GloFo) has the lead now because it has both HKMG and SOI. :twisted:

You do know that AMD is moving away from SOI? And that SOI today is a problem rather than a benefit? SOI worked great at 130nm and 90nm. Then it went all downhill.

Also its just a matter fo time before AMD entirely drops Glofo for TSMC.

 

[AtenRa]

Intel made a deliberate decision not to use SOI. It's not a straight advance -- it has both pros and cons.

Everything has its pros and cons, even HKMG, SOI (Both PD and FD) and Tri-Gate. You cant just say that GloFo or TSMC is 4 years back in lithographic process in general because Intel started HKMG 4 years ago.

 

[AtenRa]

You do know that AMD is moving away from SOI? And that SOI today is a problem rather than a benefit? SOI worked great at 130nm and 90nm. Then it went all downhill.

Also its just a matter fo time before AMD entirely drops Glofo for TSMC.

As far as i know they were talking about SOI FD for 22/20 nm.

Edit:

And that SOI today is a problem rather than a benefit

Care to provide any info about that ??

 

[ShintaiDK]

As far as i know they were talking about SOI FD for 22/20 nm.

Edit: Care to provide any info about that ??

You can also consider why AMD payed Glofo to get out of the contract. Aswell as this from AMDs CFO:
http://news.softpedia.com/news/AMD-Trinity-Architectural-Preview-Part-IV-267830.shtml

“So with respect to SOI (Silicon On Insulator), we made statements that on 28-nanometer, all of our products will be bulk,” said Mr. Seifert during the conference call regarding the Wafer Supply Agreement (WSA) amendment covering 2012.

“We said that at the 28nm node we are going to be on bulk silicon across all products, not only graphics but also CPUs. And we have made no statement beyond that. But for 28 we will be on bulk for all products,” he added.

 

[pelov]

Something I forgot to mention was uniformity. For ages it's been the biggest asset to going x86. Everything was x86, or at least pretty close to it if you count some blue side elbow cheap shots Now that's not the case for either competitor. They've started on opposite ends of the spectrum and are heading towards each other on a collision course. Whilst Intel has done wonders with lowering power consumption, it's a direct result of the fab advantage and it comes at higher cost due to Intel tax. Medfield is a great first stride but it isn't exactly revolutionary. Frankly, I think a lot of us were expecting the chip to run laps around the ARM competition but that hasn't been the case.

That's about the same advantage Intel carries against AMD in single-threaded performance In this case, though, it's nowhere near as important as the price.

Then there's Intel sucking ass GPU side again, to nobody's surprise.

This isn't to say they won't manage at the phone/tab level. They will and in some ways they already have. The one way they haven't yet managed is pricing, and competing on both price and performance is something Intel can't manage without changing their entire business model. Their business works on high profit margins due to x86 compatibility and the comfortable lead over AMD. Charging x86 tax works on servers/desktops but it won't fly in the mobile world.

I know it's been something that's been repeated many times over a period of decades, but I truly believe x86 isn't going to last much longer, regardless of the platform. HPC has taken off with purpose built coprocessors and I see this trend continuing, lessening the need for x86. The mobile world has already spoken with regards to Microsoft's bullshit and for the same reasons I don't think Intel will manage here either. Predicting the winner in the next decade is as simple as answering this question:

Can you license x86 for yourself and control its implementation and lower costs? The answer is a very clear and emphatic 'no'. The answer for ARM is a very clear and resounding 'yes'.

Just to give you some perspective,

Given that ARM's 2006 income from processor cores was approximately 60% from royalties and 40% from licenses, ARM makes the equivalent of £0.06 ($0.11) per unit shipped including both royalties and licenses. However, as one-off licenses are typically bought for new technologies, unit sales (and hence royalties) are dominated by more established products. Hence, the figures above do not reflect the true costs of any single ARM product.

 

[Dravic]

Makes going fabless sound like a pretty good deal until you realize your arch rival has the best fabs and the most money

Whats the other option, keep fighting the guy with the best fabs and most money in competition that requires the best fabs and most money?

They didn't really have a choice but to leverage additional fab resources and go after places you can better compete (Llano/Trinity). Even that will only be a limited win until Intel diverts enough $$ to fix the issue.

AMD has always been in a rough spot, but now its gotten to the point where owning a fab is so costly its driven out all smaller players. You either compete on the Intel and TSMC scale, or get out. What smaller company can afford to spend $9B a year on fab cost by themselves?

Even when AMD had a hit they couldn't sustain it do to fab cost.

 

[AtenRa]

You can also consider why AMD payed Glofo to get out of the contract. Aswell as this from AMDs CFO:
http://news.softpedia.com/news/AMD-Trinity-Architectural-Preview-Part-IV-267830.shtml

That doesn't say that SOI is not working at 32nm or it was ok up to 90nm. Give me technical info that shows that SOI doesn't working at 32nm or bellow.

One more thing, you cannot change Fabs even if they call there process 28nm. GloFos 28nm has different characteristics than TSMCs 28nm process. You will have to redesign your silicon again in order to produce it from GloFo to TSMC. That takes time, Human resources and money that AMD cannot and will not spend

If you ask me why they stopped using SOI with the 28nm GloFos process i dont know but i guess it was about cost only.

Edit: Not to mention that GloFos Fab 8 in New York will be ready for full production in 2013


http://fab2construction.com/
http://www.globalfoundries.com/manufacturing/fab8_overview.aspx

 

[ShintaiDK]

That doesn't say that SOI is not working at 32nm or it was ok up to 90nm. Give me technical info that shows that SOI doesn't working at 32nm or bellow.

One more thing, you cannot change Fabs even if they call there process 28nm. GloFos 28nm has different characteristics than TSMCs 28nm process. You will have to redesign your silicon again in order to produce it from GloFo to TSMC. That takes time, Human resources and money that AMD cannot and will not spend

If you ask me why they stopped using SOI with the 28nm GloFos process i dont know but i guess it was about cost only.

Edit: Not to mention that GloFos Fab 8 in New York will be ready for full production in 2013


http://fab2construction.com/
http://www.globalfoundries.com/manufacturing/fab8_overview.aspx

I never said SOI didnt work below 32nm. Its just not the best solution. SOI peaked long ago and its gone downhill since.

AMD knows it and thats why AMD drops SOI. SOI processes are already behind compared to bulk processes in terms of time. Not to mention SOI cost more. A large expense for no benefit.

You also seem to put Glofo and AMD too much together. After the last payment they essentially got nothing to do with one another anymore. AMD wants to get out of Glofo. They couldnt deliver enough Llano chips either.

And I dont care what Glofo builds. They are behind and getting largely irrelevant.

Guess for what foundry process AMDs 28nm chips are designed for?

Even as the extremely dedicated supporter you are. You should be happy for AMD that they can get away from Glofo and the punishment it is. Its for AMDs benefit. The CEO knows, the CFO knows. Now the fans just needs to realize it too. Glofo was a disaster relationship for AMD. They got abused so badly.

 

[AtenRa]

I never said SOI didnt work below 32nm. Its just not the best solution. SOI peaked long ago and its gone downhill since.

AMD knows it and thats why AMD drops SOI. SOI processes are already behind compared to bulk processes in terms of time. Not to mention SOI cost more. A large expense for no benefit.

You still haven't provided any info that shows that Bulk is better than SOI at GloFos 32nm. If AMD knew that SOI is not better than Bulk why they went with SOI at 32nm???

You also seem to put Glofo and AMD too much together. After the last payment they essentially got nothing to do with one another anymore. AMD wants to get out of Glofo. They couldnt deliver enough Llano chips either.

Let me guess, TSMC ??? the same TSMC that cannot supply AMD and NVIDIA 28nm GPUs ???

Guess for what foundry process AMDs 28nm chips are designed for?

I sure hope you are talking about CPU chips and not GPU. If thats what you are talking about, i hope you have a link or something else to support that.

Even as the extremely dedicated supporter you are. You should be happy for AMD that they can get away from Glofo and the punishment it is. Its for AMDs benefit. The CEO knows, the CFO knows. Now the fans just needs to realize it too. Glofo was a disaster relationship for AMD. They got abused so badly.

Why I have to be a fanboy, supporter or something else when we are talking about Fabs and Litho production ??? I dont care where AMD, Intel or NVIDA produce their Chips.

Intel went all this trouble for years, spending lots and lots of money to research Tri-Gate in order to have fully depleted transistors for 22nm and bellow and you say that AMD will stop SOI because bulk is better ??

Clearly you mistaken.

 

[zV1]

amd will be closed soon if they don't change the archirechture

 

[Cerb]

But imagine if 100 or more years from now, computers still had hardware to rename registers, deal with 64KB segments, real mode etc. Truly crazy. And that doesn't just apply to CPUs but also OSs, C-syntax programming languages etc.

You're talking about apples and oranges, though.

Everyone has hardware to rename registers. It would be stupid not to. That there should be enough registers to not need to rename was a dumb idea of early RISCs, desiring to expose the processor's workings with the ISA, which only helps in narrow worthless cases (worthless because the same things can be done w/ less registers and renaming, on other ISAs).

Very few CPUs do segmentation, mostly x86 ones. Common use for it is ancient software, hence ditching it for x86-64. Software lasts longer than hardware. x86 users today don't even get to use such features (long mode segmentation is not segmentation as we generally consider it). As 32-bit binaries die off, so will the last vestiges of it all, and good riddance.

Real mode, and equivalents, are still quite common in uCs, and aren't going anywhere any time soon. Like segmentation, we don't get to use it, haven't for many years, and nobody has looked back (I stopped in either late 1996 or early '97...most everyone else got stuck until 2001, the poor sods!).

However, 100 years from now, I would expect the very basic underpinnings of how computers are implemented to have been uprooted, leaving theoretical models for deterministic behavior of computation as the only survivors of the pre-WWII era (Lambda Calculus, FI). Everything you know today is predicated upon bandwidth and latency limitations, much of which are physical limitations. Solve them, and our assumptions of how low-level implementation of scalar computation should be done will go *poof*.

 

[Don Karnage]

Intel started shipping hkmg at 45nm, not 32nm. Intel used HKMG with Penryn aka 45nm Core 2. Or 2007. Thats 4 years.

Correct