New AMD Roadmap for 2013 (donanimhaber.com)

[ShintaiDK]

The profit margin on the arm soc can be near 0 - ARM will make $1 license which is all they need, and the companies using the chips will make it on the end product not the cpu. Intel simply can't compete with that. They can't work on that low a profit margin and they can only provide a few pre-packaged x86 chips, not license the design to allow other companies to develop the exact custom chip they want.

Look at Qualcomms margins. The entire smartphone segment is a milking machine.

Intel just want the tablet and smartphone segment. Rest it doesnt care about.

The smartphone and tablet market is what, 90%+ of ARM revenue area? Yet something like 10-15% of the amount of ARM CPUs.

 

[krumme]

Let put Pablo conspiracy theory in perspective:

If you were Otelini, would you go after a 3.5 billion gain in annual revenues from a company that is more likely to bankrupt itself out of sheer incompetence or try to break in a 50 billion market that is booming? No matter how good it would be to crush AMD, it isn't there that Intel efforts should be. The most lucrative parts are already with them.

But let's say that Intel is determined to crush AMD, is really necessary any additional efforts from Intel? Not really. Intel is committed to bring Haswell and future iterations of the Core architecture to tablets and convertibles, and more important, it wants every ultrabook to be a convertible. They have to have extreme focus on die size, power efficiency and performance, which in turn means lower ASP that will have to be compensated by smaller and more efficient dies in order to reduce COGS, exactly what Intel has been doing since Conroe but an order of magnitude bigger. So that theory that we are going to see increases of 100USD in notebooks prices is simply against their business strategy, as Intel must have a competitive package for tablets and convertibles, in other words, Intel is moving to a business model of smaller ASP, not higher. This is not a strategy to tackle AMD, this is an strategy to tackle ARM. The AMD issue solved itself in the day they launched Bulldozer.

With bulldozer they shifted to a higher ASP strategy. Intel is pretty much happy with prices where they are now, they are still getting 60%+ gross margins each quarter, while AMD went down to 38% (discounting the inventory impairment). Who needs better ASP here? The fact that they are still seen as the one that keeps Intel prices in check is more due to their inability to sell their products at the prices they want than anything else.

So when Pablo comes here stating things like "Intel's modus operandi ad vitam eternam has been to protect their amazing profits and high ASP's while causing pain for AMD in the value segment." it fails to capture, among other things, that Intel is transitioning to smaller ASP products and that AMD tried and failed to transition to high ASP products. All that is left is that Intel wants to destroy poor AMD, as if AMD wouldn't do the same if it could.

1. Its not exactly conspiracy to tell you the historical price facts. Like 3750 vs. 3770.
2. Intel is not going to crush AMD. They could easily do it now, but for reasons we both know they dont. They keep them barely alive.
3. ASP is meaningless number in itself - Intel is going for a market where there is future profit, leveraging on their enormous process competence and technologies - and a s..tload of cash.

What is interesting now is if they will succeed using the same strategy as they did against AMD. I dont think so:

1. Its different segment, and customers, that selling fx. the new ssd type to b2b. It means they can not use their sales capacity the same way as they do for their existing portfolio.
2. They are competing against Samsung and TSMC. Who wants to do that - and they dont have a brand advantage like Apple does doing so.

But Intel have proved they could go into new markets before. So its going to be a hell of a fight.

Bing !

 

[mrmt]

1. Its not exactly conspiracy to tell you the historical price facts. Like 3750 vs. 3770.!

I don't get it.

If you are talking about the minute difference in performance but significant difference in price, I don't see why the fuss. It is always like that in every other market around. Be it in clothes, cars, sound, phones, tablets, airplanes, anything you pay more for each small increase you go in the value ladder. Even AMD did that in their golden years.

2. Intel is not going to crush AMD. They could easily do it now, but for reasons we both know they dont. They keep them barely alive.

I don't really buy this theory that Intel wants/needs to keep AMD alive. In fact, Intel is being quite aggressive since AMD got the upper hand in the early 2000's.

First before Conroe they undercut AMD on pricing, then after they launched it pricing was more adequated to grab a lot of market share than to shoot their margins to the stars, this in a moment that AMD was extremely vulnerable because of the ATI acquisition.

Then in 2008 AMD stock went to 1.82, AMD was drowing in debt and would not live through 2009 for sure. When the spin off happened they weren't that constructive, and they only conceded to amend the agreement after AMD agreed to drop charges that would hurt far more if they were pushed until the end.

When Intel launched westmere it started to undermine AMD on its traditional strongholds, the bottom market. Intel kept pushing with SNB and now with IVB the bottom market is becoming Intel field too, as AMD chips are too big to be competitive there. The Brazos niche is going to suffer with cheap IVB celerons and Silvermont too, so AMD gets no relief here too.

To me this looks like Intel trying to crush AMD, but not using its financial muscle, but using its R&D muscle to push AMD down into the value chain to the point that the company ceases not only to be competitive but to be viable. That way they can present the authorities with a de facto monopoly, but not one that was result of their financial or comercial practices, but one that resulted from offering consumers the best products. (not that I always agree with it).

 

[podspi]

There seems to be a lot of emotions into this. I set strategy for a small company, at not until this discussion have it occurred to me, that some people find it suspect, to sell the same product with 50% discount when competition enters. I have no problem selling at a loss, to keep compettion at bay.

Is there something wrong with that?

How do you think it is to manage a business? Its often a brutal fight, that turns worse the bigger the business is. Go read fx. the transcript and mails between Otellini and Michael (Dell). This is how its done behind the rosy curtains.

I won't get into a moral argument about whether it is right or wrong, but at least in the United States it is illegal. Competition is brutal, but like warfare there are still rules. You might be able to get away with not following them, or you might not.

 

[blckgrffn]

Not this time - they will come up with some pre packaged products at highish prices which is all they know how to do and all they can do if they want to make the profit margins the share holders require. All the phone/tablet companies will take a look and say "I'll keep my custom arm soc that might not be quite as good in some areas but does exactly what I want it too, gives me complete control over design and costs peanuts".

The profit margin on the arm soc can be near 0 - ARM will make $1 license which is all they need, and the companies using the chips will make it on the end product not the cpu. Intel simply can't compete with that. They can't work on that low a profit margin and they can only provide a few pre-packaged x86 chips, not license the design to allow other companies to develop the exact custom chip they want.

We'll see, I suppose. If Intel thinks it is in their best, long term interest to make a big dent in the mobile market, I trust they will find a way - even if they compromise margins *a bit* in order to stay in the target market. Best in class chips on very small dies... and Intel Inside sells. Those are the two reasons I think Intel is going to stay relevant in the mobile era.

Right now we see their dominance in the desktop, laptop and server markets because they make such good stuff we buy it.

If Intel can pre-package an Atom with significant performance and power advantages over everything but maybe the most boutique in-house ARM designs, what is the point? Design and integration teams aren't free.

Apple and Samsung might have the cash flow and volume to make it happen and nvidia has their own reasons for ARM investment, but other players might really like a little blue sticker to distance themselves from those competitors if they are relying on more traditional ARM cpus that don't compare favorably. If your SoC costs "peanuts" but puts you at significant competitive disadvantage it might be costing you a helluva lot more than is immediately obvious.

 

[Dribble]

Intel do so well because they locked down the pc market with x86 - every pc has to have an x86 processor and an x86 motherboard, and intel completely control x86 so they make what they like and everyone else just has to use it, and pay whatever Intel asks. AMD was mostly Intels puppet to keep monopolies people at bay, but it could still do ok as it was the only other company allowed into the walled x86 garden of delights.

This new ARM emergence has finally allowed companies to break out of the x86 stranglehold. Now they can license arm and make what they like - no longer are they subject to the whims of Intel. Now they can spec whatever they like at whatever price/chip they want - they license the various bits at very low prices, and get one of a number of fabs to build it. I very much doubt they want to go back to only being able to produce the machines intel let them.

Intel can obviously still compete in several ways - they have the best fabs for a start, but their shareholders have been spoilt for a long time by huge margins that Intel could generate with it's near monopoly and efficient use of fabs. Now they've lost control they simply can maintain those margins, and the company is so huge that they are in trouble as costs go down - if in a few years the cpu/mb goes from $150 (approx price Intel was able to maintain them at) to $15 (approx price for a tablet soc) then even if Intel make all of them there's only a fraction of the money out there to be made.

 

[boxleitnerb]

Even Charlie says now that Kaveri is coming in 2014, apparently because it was/is being "re-evaluated" to better position it against Haswell.

 

[mrmt]

Even Charlie says now that Kaveri is coming in 2014, apparently because it was/is being "re-evaluated" to better position it against Haswell.

Yes, it will be interesting to see a chip better positioned against Haswell by the time Broadwell arrives. /sarcasm

http://semiaccurate.com/2012/11/06/amds-kaveri-apu-slips-again-2014-now/

More recently, SemiAccurate's moles have come back and said that Kaveri has slipped yet again to 2014 if it is still alive. Big if. That would put the chip 18 months late best case, and then you have to ask yourself why not just skip it?

 

[cytg111]

Even Charlie says now that Kaveri is coming in 2014, apparently because it was/is being "re-evaluated" to better position it against Haswell.

reevaluated to be what ? Harder, better, faster, stronger? .. wtf, i smell crap. surely its not gaining position by being a year late ..

 

[Idontcare]

Yes, it will be interesting to see a chip better positioned against Haswell by the time Broadwell arrives. /sarcasm

Right now we are in Q4/12.

There is no way in hell AMD can (1) alter Kaveri in any meaningful way that markedly improved performance over the current design, (2) get the Kaveri+ design through tapeout, (3) get first-silicon samples back from the fab, (4) iterate through the standard silicon verification and debug cycle, (5) respin whatever fixes are needed, (6) get second-silicon samples back from the fab, (7) iterate through the standard silicon verification and debug cycle, and (8) ramp the fab for 3 months to build inventory as needed in advance of product release to market...all within the span of 12-15 months, its just not possible.

It would be possible if step (1) above were not involved, but the whole premise of the argument is that step (1) is involved, which puts the timeline on a 2015 release date, not 2014.

Pie in the sky stuff is going on if people are being told to expect a Kaveri+ design coming out in 2014 based on management decisions being made in Q4/12.

 

[pelov]

Does anyone know of any high power nodes available to AMD post 28nm? If there's an alternative somewhere at 20nm then that could explain a delayed Kaveri being primed for another node. Richland, or Piledriver refresh APU, is on GloFo's 28nm. I highly doubt AMD is going to offer a Kaveri on a 28nm bulk node to face off against a 14nm Broadwell.

 

[mrmt]

Pie in the sky stuff is going on if people are being told to expect a Kaveri+ design coming out in 2014 based on management decisions being made in Q4/12.

Charlie article mentions a 18 months, not a twelve months delay, which puts Kaveri in H214, not H114. If the decision was taken one or two months ago and now leaked, that would tightly fit a schedule like the one you brought us.

But still, something is fishy here. If they were just expecting another node or reached the conclusion that Kaveri won't be relevant, why not just skip steamroller and go straight to excavator? To me, this seems like problem in their R&D pipeline. They hit something bad there, such as lack of engineering resources to develop the chip or a thermal barrier like Intel reached with Prescott.

 

[mrmt]

Does anyone know of any high power nodes available to AMD post 28nm? If there's an alternative somewhere at 20nm then that could explain a delayed Kaveri being primed for another node. Richland, or Piledriver refresh APU, is on GloFo's 28nm. I highly doubt AMD is going to offer a Kaveri on a 28nm bulk node to face off against a 14nm Broadwell.

If R&D pipeline is on track why not just skip Kaveri and go straight to excavator? Excavator was due in 2014 anyway.

 

[boxleitnerb]

reevaluated to be what ? Harder, better, faster, stronger? .. wtf, i smell crap. surely its not gaining position by being a year late ..

Beats me. If you have a product that is somehow competitive, just release it and work harder on the successor.

Right now we are in Q4/12.

There is no way in hell AMD can (1) alter Kaveri in any meaningful way that markedly improved performance over the current design, (2) get the Kaveri+ design through tapeout, (3) get first-silicon samples back from the fab, (4) iterate through the standard silicon verification and debug cycle, (5) respin whatever fixes are needed, (6) get second-silicon samples back from the fab, (7) iterate through the standard silicon verification and debug cycle, and (8) ramp the fab for 3 months to build inventory as needed in advance of product release to market...all within the span of 12-15 months, its just not possible.

It would be possible if step (1) above were not involved, but the whole premise of the argument is that step (1) is involved, which puts the timeline on a 2015 release date, not 2014.

Pie in the sky stuff is going on if people are being told to expect a Kaveri+ design coming out in 2014 based on management decisions being made in Q4/12.

It could work if this whole process started some time ago.

 

[Idontcare]

Does anyone know of any high power nodes available to AMD post 28nm? If there's an alternative somewhere at 20nm then that could explain a delayed Kaveri being primed for another node. Richland, or Piledriver refresh APU, is on GloFo's 28nm. I highly doubt AMD is going to offer a Kaveri on a 28nm bulk node to face off against a 14nm Broadwell.

If you believe the marketing materials, ya know the same ones that would have had you believing that GloFo would be shipping 28nm last year instead of next year, then GloFo will be shipping gobs of 14nm-XM chips in 2014.

Charlie article mentions a 18 months, not a twelve months delay, which puts Kaveri in H214, not H114. If the decision was taken one or two months ago and now leaked, that would tightly fit a schedule like the one you brought us.

But still, something is fishy here. If they were just expecting another node or reached the conclusion that Kaveri won't be relevant, why not just skip steamroller and go straight to excavator? To me, this seems like problem in their R&D pipeline. They hit something bad there, such as lack of engineering resources to develop the chip or a thermal barrier like Intel reached with Prescott.

The "something fishy" part is beginning to look suspiciously like a total cancellation is actually in the works but the memo is itself is being intentionally delayed for a bit. Won't surprise me one bit if the other shoe takes a solid 3 months to drop. Big restructuring is always rolled out in stages so the changes can be absorbed without killing existing sales like a guillotine.

It could work if this whole process started some time ago.

The decision could not have been made all that long ago, if it was then AMD placed itself at risk of being sued by shareholders for having misled them in terms of their analyst roadmaps. Roadmaps can change, but they must be reflective of managements intentions at the time they are divulged. You aren't allowed to lie or commit subterfuge with roadmaps. So the changes could not be any more older than the day after the most recent analyst meeting.

But still, mrmt is correct in that this could still be on H2/14 timeline. Which would put them on track to compete with 14nm broadwell. Uhg. I hope the "something fishy" is that it is completely cancelled and instead they are diverting resources to pulling in 14nm-XM excavator for a 2014 release.

14nm-XM excavator would at least have a shot at competing with the performance/watt and absolute performance of a Haswell (and possibly a Broadwell?) but it needs to come out in 2014, not 2015 or 2016.

 

[Keysplayr]

n/m

 

[jones377]

Answering my own questions about "Radeon Cores 2.0" after a bit of googling.

http://www.amd.com/us/products/desktop/processors/a-series/Pages/a-series-pib.aspx

Here AMD themselves refer to the GPU in Trinity as Radeon Cores 2.0. In light of this, it really looks like Richland is nothing more than Trinity 2.0. I also now really doubt it will move to a 28nm bulk process but will instead stay on 32nm SOI.

Richland vs Haswell will not be pretty. Already AMD positions Trinity vs Core i3 IB's. What will they position Richland against? Haswell Celerons?

 

[mrmt]

14nm-XM excavator would at least have a shot at competing with the performance/watt and absolute performance of a Haswell (and possibly a Broadwell?) but it needs to come out in 2014, not 2015 or 2016.

Isn't 14nm-XM process tuned for SoCs, meaning that the process is aiming low power, low frequency, high density chips, exactly the opposite of the design premises of Bulldozer chips?

Isn't that same process based on their 20nm process, a process that still hasn't manufactured a a single commercial chip?

 

[blckgrffn]

Answering my own questions about "Radeon Cores 2.0" after a bit of googling.

http://www.amd.com/us/products/desktop/processors/a-series/Pages/a-series-pib.aspx

Here AMD themselves refer to the GPU in Trinity as Radeon Cores 2.0. In light of this, it really looks like Richland is nothing more than Trinity 2.0. I also now really doubt it will move to a 28nm bulk process but will instead stay on 32nm SOI.

Richland vs Haswell will not be pretty. Already AMD positions Trinity vs Core i3 IB's. What will they position Richland against? Haswell Celerons?

How much absolute performance is Haswell going to bring to the table? Perf/Watt, yes... but it is clear we are no longer in performance era for x86 desktop computing. In terms of "how fast can CPU A get a task done vs CPU B" it might be pretty much status quo, just at a huge efficiency deficit.

At the same time, for the majority desktop users, that perf/watt difference is not a big deal either in real world costs or perception. You don't walk into costco and see "45W CPU Inside!"

AMD vs i3 might persist for quite a while.

 

[boxleitnerb]

10+% more IPC, maybe 10% higher clocks and improved hyperthreading, significantly improved GPU...CPU-wise Richland will lose so bad it won't even be funny. GPU-wise there might be a tie or a small advantage for AMD.

 

[mrmt]

At the same time, for the majority desktop users, that perf/watt difference is not a big deal either in real world costs or perception. You don't walk into costco and see "45W CPU Inside!"

It makes a hell of a diference once you factor notebooks, all in one and tablets as you need less cooling. With less cooling needs you can have smaller costs, better form factors, or both, exactly things that can be make or break for a given SKU.

 

[Idontcare]

Isn't 14nm-XM process tuned for SoCs, meaning that the process is aiming low power, low frequency, high density chips, exactly the opposite of the design premises of Bulldozer chips?

Isn't that same process based on their 20nm process, a process that still hasn't manufactured a a single commercial chip?

GloFo is definitely pushing the low-power aspects of the 14nm-XM as a priority over that of emphasizing the benefits it will bring to high-power designs but my gut feeling on that is that they are doing this because the low-power characteristics of their 20nm node "sucks wind" to use some technical industry jargon

They are promoting it as a means of convincing would be low-power 20nm customers to not abandon ship based on the early silicon results they are getting on 20nm shuttle wafers.

But 14nm-XM will bring benefits to the high performance ICs as well, its just not being emphasized by GloFo at this time because they have a mobile issue with their 20nm.

Checkout slides 14 an 16 of this GloFo pdf.

Note the Vdd vs fmax curve, that will be taken advantage of by anyone making 20nm CPU's at GloFo:

Note the "Optimized CPU Solutions" aspect:

Look at it from this perspective - if GloFo's 14nm-XM does nothing to provide better performance for the high-power MPU industry then that means GloFo has abandoned the high-power MPU segment entirely because you can't just skip it for a node or two and not lose customers to TSMC who will be offering it.

 

[blckgrffn]

It makes a hell of a diference once you factor notebooks, all in one and tablets as you need less cooling. With less cooling needs you can have smaller costs, better form factors, or both, exactly things that can be make or break for a given SKU.

Desktop users. It's not like that segment is going away overnight.

I agree with you on the other segments, so does Intel, obviously.

We are all going to get "good enough" desktop parts for the foreseeable future. Yay.

10+% more IPC, maybe 10% higher clocks and improved hyperthreading, significantly improved GPU...CPU-wise Richland will lose so bad it won't even be funny. GPU-wise there might be a tie or a small advantage for AMD.

If Haswell is to Ivy Bridge as IVB was to SB in terms in performance in 99% of the use cases, that's a pretty big snoozer. Not that reducing power consumption and mainboard costs is something to overlook, its just boring in the context of this forum.

And it means that if Intel doesn't change up their product structure much (ie, i3, i5, Pentiums, etc.) APU vs APU is going be close enough to be immaterial to people buying desktop PCs.

 

[ShintaiDK]

How much absolute performance is Haswell going to bring to the table? Perf/Watt, yes... but it is clear we are no longer in performance era for x86 desktop computing. In terms of "how fast can CPU A get a task done vs CPU B" it might be pretty much status quo, just at a huge efficiency deficit.

At the same time, for the majority desktop users, that perf/watt difference is not a big deal either in real world costs or perception. You don't walk into costco and see "45W CPU Inside!"

AMD vs i3 might persist for quite a while.

For (recompiled) SSE/AVX tasks alone we talk a potential 50-60% for example. All depends on the code. Other times it might only be 10%. But Linpack for example will benefit greatly.

 

[blckgrffn]

For (recompiled) SSE/AVX tasks alone we talk a potential 50-60% for example. All depends on the code. Other times it might only be 10%. But Linpack for example will benefit greatly.

That's great for my crunchers, but what is that for most people?

To be clear, I have a number of dedicated Intel crunching machines and I can appreciate this some - but most distributed computing tasks don't even use AVX yet.