ambidextrous computing; AMD project skyla..skybridge!

[witeken]

The troubling thing (or Intel) is that the rest of the ARM-based marketspace is doing just fine without x86 and without Intel's process node lead.

Because a process node lead doesn't exist in the mobile market... yet. Intel is now introducing their process lead into both markets, suddenly improving SoC tech by 2 years. A company that takes a 22nm SoC will be able to make a smartphone that is performing at a level that would otherwise only have been possible 1-2 years later. Because no one wants their SoCs to use such outdated tech, the market will automatically move to Intel (in theory). This will be more pronounced at 14nm, which TSMC is only going to match about 3 years later.

 

[Idontcare]

Because a process node lead doesn't exist in the mobile market... yet. Intel is now introducing their process lead into both markets, suddenly improving SoC tech by 2 years. A company that takes a 22nm SoC will be able to make a smartphone that is performing at a level that would otherwise only have been possible 1-2 years later. Because no one wants their SoCs to use such outdated tech, the market will automatically move to Intel (in theory). This will be more pronounced at 14nm, which TSMC is only going to match about 3 years later.

As a process node development engineer, I love the idea that such a future will come to pass, thus vindicating the value-add of my particular choice of career.

Unfortunately this is Intel we are talking about, a company whose history is full of strategy and well spoken plans about diversification but equally full of failure in literally every regard outside their shining moment of taking over the desktop PC market and then skewering big-iron in the server space.

We all want Intel to succeed because that would mean all of our devices would suddenly see a 2-3yr leap in capability. Apple did great in the desktop and laptop markets when they jumped from powerpc to x86.

But a healthy dose of skepticism is warranted here. Intel had viable plans to diversify into HDTV(2005), mobile phones (2006), and discrete GPUs (2007) in recent years past. And they were all failed ventures. What Intel lacks is a proven history of leveraging their process technology coupled with an IC that creates a market opportunity outside of their established duopoly space.

I have every (mathematical) reason to expect great things to come of Intel in the coming 14nm and 10nm product cycles, but history is not on their side when speaking to their ability to effectively manage the process of having the rubber meet the road in all things non-traditional "x86" marketspace.

But every consumer on the planet stands to benefit if Intel does finally do the hitherto impossible and actually break into new marketspace; because the coffers of Apple, Qualcomm, and Samsung are filled with profits gleaned from consumers past and that speaks to their plans for the future as well. More competition is in order, and Intel stands to be something different in a world of all things ARM.

 

[AtenRa]

Global Server Market revenue in 2013 was ~52B USD
Forecast for 2017 is 55-60B
ARM is targeting 10% Server market share for 2017, that is 5.5-6B USD
AMD with 40% ARM server market share could have 2.2-2.4B revenue. That is from ARM server revenue alone per year, not bad.
With 40% margins ( i know it is low for server parts) they could make 880-960M profit.

 

[witeken]

As a process node development engineer, I love the idea that such a future will come to pass, thus vindicating the value-add of my particular choice of career.

Unfortunately this is Intel we are talking about, a company whose history is full of strategy and well spoken plans about diversification but equally full of failure in literally every regard outside their shining moment of taking over the desktop PC market and then skewering big-iron in the server space.

We all want Intel to succeed because that would mean all of our devices would suddenly see a 2-3yr leap in capability. Apple did great in the desktop and laptop markets when they jumped from powerpc to x86.

But a healthy dose of skepticism is warranted here. Intel had viable plans to diversify into HDTV(2005), mobile phones (2006), and discrete GPUs (2007) in recent years past. And they were all failed ventures. What Intel lacks is a proven history of leveraging their process technology coupled with an IC that creates a market opportunity outside of their established duopoly space.

I have every (mathematical) reason to expect great things to come of Intel in the coming 14nm and 10nm product cycles, but history is not on their side when speaking to their ability to effectively manage the process of having the rubber meet the road in all things non-traditional "x86" marketspace.

But every consumer on the planet stands to benefit if Intel does finally do the hitherto impossible and actually break into new marketspace; because the coffers of Apple, Qualcomm, and Samsung are filled with profits gleaned from consumers past and that speaks to their plans for the future as well. More competition is in order, and Intel stands to be something different in a world of all things ARM.

I understand your points, but it's ironic that while you are claiming history is not on Intel's side in terms of leveraging their process node, there's a nice article about Intel's history on Seeking Alpha. Intel is literally going all-in on the mobile and even IoT markets, so I think history really is on their side.

Brian Krzanich extensively went into the reason they have for going into those markets, in the Q1 conference call: (TL;DR in bold)

Remember, Intel has two assets. We have our silicon technology, but we also have our architecture. And one of the things an OEM gets when they build with Intel technology is that they can go into any OS and they can build a single platform and move that on to Chrome, on to Android, on to Windows. And that's a very unique capability that we provide to OEMs for flexibility.

So, we believe with a product like SoFIA, as we bring that into the market next year, we can absolutely compete in those spaces and make money. You're probably not going to make as much revenue dollars and as much margin dollars as the PC business, but we think this is still critical. And it's critical for a variety of reasons. Part of it is simply the scale. You want to have those units. You want to have a presence in all areas of computing.

And the second one is developer attention. You want developers creating new products, doing innovation on your architecture. This is a space that's got innovation. We are going to bring some of that innovation to this market. You're going to see some tablets as you go into the end of this year.

We showed them at CES, some of the highlights where you have 3D cameras, you have perceptual computing capabilities for gaming. All of those kinds of things can change the tablet market, along with the PC market.

So, we believe that we can bring a lot of the innovation that we do in the PC down into the tablet space. And again, that keeps the developers developing and interested in our platform. I think for all of those reasons, we want to be in this space and we will be in this space from now on.

Source.

 

[mrmt]

But a healthy dose of skepticism is warranted here. Intel had viable plans to diversify into HDTV(2005), mobile phones (2006), and discrete GPUs (2007) in recent years past. And they were all failed ventures. What Intel lacks is a proven history of leveraging their process technology coupled with an IC that creates a market opportunity outside of their established duopoly space.

There is a subtle but significant difference here. These ventures you mentioned would be nice nice additions to Intel portfolio, but by no means were they "do or die" situations for Intel. GPU were relatively small, HDTV was very low margin and phones were deemed a small niche. Albeit profitable, those "nice to have" things were competing against with Intel mainstream product portfolio, in a time where this mainstream portfolio was growing at a healthy doses.

With mobile today, things are a little different. Intel *needs* to break in otherwise their entire business model, mainstream products included, will be impacted. Mobile is seeing to getting a sizable pie of Intel crack teams and resources, something that these products you mentioned were not even close to get. When would you expect Intel to throw away 1 billion to break into a market, or manufacture a x86 SoC in an external foundry, or ramp up a bleeding edge node with an Atom product? Not under Otelini in 2006.

Those are decisions taken by a company which wants things to happen *fast*, is willing to pay for it, and is willing to stomach the compromises. This time I expect Intel to try or die trying this time, something along of the line of Nvidia with Tegra.

 

[witeken]

Just ten minutes ago I read this article on ExtremeTech. That article -- and ARM -- is way too optimistic about ARM's reign, saying "low- and mid-range smartphone and tablet markets to more than double over the next five years, resulting in almost 2.5 billion total mobile device shipments by 2018… and they’ll (probably) all use ARM chips." Unless TSMC does some magic tricks in the coming years, 2018's computing landscape will be dominated by 7nm x86 built on 450mm wafers.

 

[ShintaiDK]

Just ten minutes ago I read this article on ExtremeTech. That article -- and ARM -- is way too optimistic about ARM's reign, saying "low- and mid-range smartphone and tablet markets to more than double over the next five years, resulting in almost 2.5 billion total mobile device shipments by 2018… and they’ll (probably) all use ARM chips." Unless TSMC does some magic tricks in the coming years, 2018's computing landscape will be dominated by 7nm x86 built on 450mm wafers.

That does sound unlikely. The trend is to replace regular phones with smartphones. But to ship 2.5B, they would have to ship a billion more phones than currently. That means alot of people around the world need to earn much more money to pay for it. I dont see that happening.

 

[teejee]

That does sound unlikely. The trend is to replace regular phones with smartphones. But to ship 2.5B, they would have to ship a billion more phones than currently. That means alot of people around the world need to earn much more money to pay for it. I dont see that happening.

Budget smartphones are soon reaching 20$.
http://www.anandtech.com/show/7993/...hones-in-2018-20-smartphones-coming-this-year
And I don't believe that Intel will sell 2-3$ SOC's with 7nm/450mm tech in 2018. They want better return for their money. I think this is a market they neither will nor can take.

 

[Ajay]

Ah, OK, I recommend you don't use Nvidia as an indicator of TSMC's 20nm HOL (health of line) because there are some things going on which are causing that and it doesn't have much to do with TSMC or its 20nm situation.

Nvidia is having issues adjusting to the reality that the GPU stopped being the fab filler they thought it would always be because TSMC found that the mobile business (anything ARM) is ridiculously higher volume than the discrete GPU business and the customers in mobile to be far less "violently disagreeable" in public.

Yeah, in retrospect, Jen-Hsun Huang really screwed the pooch on that one. But with the rumor that Apple is going back to Samsung for 14nm, IIRC, NV may get earlier access to TSMC's 16FF wafers than they are for 20nm.

 

[Ajay]

Budget smartphones are soon reaching 20$.
http://www.anandtech.com/show/7993/...hones-in-2018-20-smartphones-coming-this-year
And I don't believe that Intel will sell 2-3$ SOC's with 7nm/450mm tech in 2018. They want better return for their money. I think this is a market they neither will nor can take.

Well, Intel might be able to participate in the lower end smart phone with SoFIA on an n-1 node (or maybe even with an older x86 SoC) - but the "rock bottom" smartphone market would likely be something they will avoid. "Interesting" times for SoC developers and fabs - market inflection points are hard on the nerves!

 

[cbn]

Budget smartphones are soon reaching 20$.
http://www.anandtech.com/show/7993/...hones-in-2018-20-smartphones-coming-this-year
And I don't believe that Intel will sell 2-3$ SOC's with 7nm/450mm tech in 2018. They want better return for their money. I think this is a market they neither will nor can take.

Well, Intel might be able to participate in the lower end smart phone with SoFIA on an n-1 node (or maybe even with an older x86 SoC) - but the "rock bottom" smartphone market would likely be something they will avoid. "Interesting" times for SoC developers and fabs - market inflection points are hard on the nerves!

With regard to "rock bottom smartphone", I actually hope (in some ways) Intel resists improving silvermont atom.

For a cheap x86 phone, silvermont atom should be fine for many years......provided it is inexpensive enough.

(With that mentioned, bring on the new Goldmont atom cpu core and whatever midrange cpu core Intel is developing......but make the "good enough" low end (ie, Silvermont atom) even more affordable if possible)

 

[Idontcare]

Yeah, in retrospect, Jen-Hsun Huang really screwed the pooch on that one. But with the rumor that Apple is going back to Samsung for 14nm, IIRC, NV may get earlier access to TSMC's 16FF wafers than they are for 20nm.

Not going "back" to Samung for 14nm. Apple has long wanted a dual-source solution to its foundry needs. Understandably so. At TI we maintained concurrent contracts with three foundries for this reason.

For 14nm they will use TSMC and GF. In order to use GF they needed GF to have a compelling 14nm process, ergo the licensing deal.

There is an interesting back story to this, and believe it or not Al Gore was involved (not joking!).

 

[Idontcare]

I understand your points, but it's ironic that while you are claiming history is not on Intel's side in terms of leveraging their process node, there's a nice article about Intel's history on Seeking Alpha. Intel is literally going all-in on the mobile and even IoT markets, so I think history really is on their side.

Brian Krzanich extensively went into the reason they have for going into those markets, in the Q1 conference call: (TL;DR in bold)

There is a subtle but significant difference here. These ventures you mentioned would be nice nice additions to Intel portfolio, but by no means were they "do or die" situations for Intel. GPU were relatively small, HDTV was very low margin and phones were deemed a small niche. Albeit profitable, those "nice to have" things were competing against with Intel mainstream product portfolio, in a time where this mainstream portfolio was growing at a healthy doses.

With mobile today, things are a little different. Intel *needs* to break in otherwise their entire business model, mainstream products included, will be impacted. Mobile is seeing to getting a sizable pie of Intel crack teams and resources, something that these products you mentioned were not even close to get. When would you expect Intel to throw away 1 billion to break into a market, or manufacture a x86 SoC in an external foundry, or ramp up a bleeding edge node with an Atom product? Not under Otelini in 2006.

Those are decisions taken by a company which wants things to happen *fast*, is willing to pay for it, and is willing to stomach the compromises. This time I expect Intel to try or die trying this time, something along of the line of Nvidia with Tegra.

You both make valid points. Intel's motivation is different now versus 10yrs ago, and their opportunity is different now as well (given the node gap that 14nm and 10nm are going to create).

By the numbers, the future is theirs for the taking. No disagreement from me there. Provided they don't hire Elop...

 

[Ajay]

Not going "back" to Samung for 14nm. Apple has long wanted a dual-source solution to its foundry needs. Understandably so. At TI we maintained concurrent contracts with three foundries for this reason.

For 14nm they will use TSMC and GF. In order to use GF they needed GF to have a compelling 14nm process, ergo the licensing deal.

There is an interesting back story to this, and believe it or not Al Gore was involved (not joking!).

Oh interesting! I guess the rumor I read was way off (some Mac site). So, yeah, the licensing deal really makes sense now. Actually, I think you or mrmt may have already mentioned this

So, Al Gore invented the A9 Seriously, I love to hear the story, unless it was told in confidence.

By the numbers, the future is theirs for the taking. No disagreement from me there. Provided they don't hire Elop...

I hope that's a joke!

 

[jpiniero]

So, Al Gore invented the A9 Seriously, I love to hear the story, unless it was told in confidence.

Gore is on Apple's Board, so him working on a deal on behalf of Apple doesn't seem that extraordinary.

 

[Idontcare]

Gore is on Apple's Board, so him working on a deal on behalf of Apple doesn't seem that extraordinary.

True, the extraordinary part here is that this is a case where a board member (any board member) was actually "working"...that is an event not often witnessed!

 

[Vesku]

More signs of interest in ARM for servers:

'A part of Nvidia's Financial Q1 2015 conference call Q&A session included some questions about micro servers, whether or not the 64-bit Tegra K1 can make it into the GRID market.

Nvidia's CEO Jen-Hsun Huang was a straight shooter saying that Nvidia is “seeing a lot of interest in putting something like Tegra in micro servers,” but he added a caveat: “one step at a time, one step at a time."'

http://fudzilla.com/home/item/34705-nvidia-might-enter-arm-micro-server-market

 

[cbn]

More signs of interest in ARM for servers:

'A part of Nvidia's Financial Q1 2015 conference call Q&A session included some questions about micro servers, whether or not the 64-bit Tegra K1 can make it into the GRID market.

Nvidia's CEO Jen-Hsun Huang was a straight shooter saying that Nvidia is “seeing a lot of interest in putting something like Tegra in micro servers,” but he added a caveat: “one step at a time, one step at a time."'

http://fudzilla.com/home/item/34705-nvidia-might-enter-arm-micro-server-market

Remember this one:

http://www.anandtech.com/show/7668/gigabyte-at-ces-2014-brix-max-sff-nas-running-android

An android NAS (believe it or not)

I didn't even know Android was capable of being used as a NAS OS.

 

[Arachnotronic]

Not going "back" to Samung for 14nm. Apple has long wanted a dual-source solution to its foundry needs. Understandably so. At TI we maintained concurrent contracts with three foundries for this reason.

For 14nm they will use TSMC and GF. In order to use GF they needed GF to have a compelling 14nm process, ergo the licensing deal.

There is an interesting back story to this, and believe it or not Al Gore was involved (not joking!).

IDC,

This whole deal doesn't make sense. Why would Samsung - which currently gets the wafer margins for Apple's business - just license out its process to GloFo for likely some upfront payment and probably some royalty stream?

Why wouldn't Samsung just tell Apple to go pound sand and go all-in at TSMC if it doesn't want to build chips at Samsung? Why let Apple have its cake and eat it too (getting the process recipe, but not supporting Samsung)?

 

[jpiniero]

IDC,

This whole deal doesn't make sense. Why would Samsung - which currently gets the wafer margins for Apple's business - just license out its process to GloFo for likely some upfront payment and probably some royalty stream?

Why wouldn't Samsung just tell Apple to go pound sand and go all-in at TSMC if it doesn't want to build chips at Samsung? Why let Apple have its cake and eat it too (getting the process recipe, but not supporting Samsung)?

Maybe going back to Samsung for 14 nm was part of the deal.

 

[Arachnotronic]

Maybe going back to Samsung for 14 nm was part of the deal.

IDC has stated specifically that Apple will not go back to Samsung for 14nm - it's TSMC and GloFo.

 

[Mr. Pedantic]

I remember several months ago everyone was doom and gloom about AMD's future in x86, and there was even an Anandtech podcast where Anand and Brian were discussing AMD's options. Where did that go and how did it pan out? I notice AMD's still here...anything they had to give up or change? Or is it all just business as usual?

they were asked why not just leave x86 for arm fulltime...they gave a PR-ish response but I believe they just like the abuse.

Better the devil you know than the devil you don't?

In x86 AMD "only" has to compete with Intel. In ARM they have to compete with Qualcomm, Nvidia, Samsung, Apple, AND Intel, as well as trying to establish themselves in a market where they don't really have a foothold yet.

 

[Vesku]

Better the devil you know than the devil you don't?

In x86 AMD "only" has to compete with Intel. In ARM they have to compete with Qualcomm, Nvidia, Samsung, Apple, AND Intel, as well as trying to establish themselves in a market where they don't really have a foothold yet.

I'm sure that's a factor but it's a somewhat silly question even if the questioner was themselves' convinced AMD would make a killing with ARM products. AMD does not have some hidden pile of money to make up the lost x86 revenue while they ramp up ARM.

 

[Idontcare]

IDC,

This whole deal doesn't make sense. Why would Samsung - which currently gets the wafer margins for Apple's business - just license out its process to GloFo for likely some upfront payment and probably some royalty stream?

Why wouldn't Samsung just tell Apple to go pound sand and go all-in at TSMC if it doesn't want to build chips at Samsung? Why let Apple have its cake and eat it too (getting the process recipe, but not supporting Samsung)?

Business savvy on Samsung's part.

Tell Apple to go pound sand and what have you done? Apple gives all that revenue to TSMC, which feeds back into TSMC R&D thereby making Samsung's biggest headache in the foundry space an even bigger one down the road.

But give Apple what it wants, a dual-foundry source solution and guarantee that at least some of foundry revenue generated by Apple is going to be split such that TSMC isn't getting 100% of it.

In the meantime, even if the plan of record is to fill the Malta fab with Apple chips thanks to Samsung's 14nm process, Samsung knows it has the opportunity to score some of those Apple wafer starts down the road should GF drop the ball in terms of yields, capacity, cycle-time, wafer costs, etc when it comes time to ramp HVM.

In other words they leave the door open for themselves to maybe bag a non-zero percent of Apple's 14nm wafer starts, but even if they don't get a single wafer order from Apple they are ensuring that TSMC is not bagging 100% of the opportunity.

It is a "hedge your bets" type of a play. And they (Samsung) know 10nm is going to play out in the exact same way (if Samsung decides to enable it) because they know GF has nothing viable in the works for 10nm as well.

Keep the door open for wafer orders to flow back to your fabs, keep your biggest competitor from becoming all the bigger, and be the puppet-master in defining the future of your other competitor. Seems like a worthwhile venture.

However, it is not like this was Samsung's dream solution. To pull it off Mr Gore had to manage the face-to-face personally. Prior to that, Samsung was perfectly willing to tell Apple to go pound sand and just take the risk of creating a bigger monster out of TSMC.

Expect more consolidation from GF, they desperately need an in-house R&D. I haven't heard anything about this, so this is just my opinion, but I would not be surprised if they bought UMC and the IBM fab assets to cobble something together that would give them a shot at being independent again for 7nm.

Edit: just realized this has absolutely nothing to do with the thread title, I apologize to everyone for contributing to the thread's derailment, my bad No more of this Apple talk, let's get back to the thread's topic.

 

[sirroman]

Isn't this talk worthy of a spin-off?