ambidextrous computing; AMD project skyla..skybridge!

[witeken]

Everybody knows that ARM is an inherently more efficient ISA than x86. Don't take my word. Microprocessor design veterans like Jim Keller are of that view

http://techreport.com/review/26418/amd-reveals-k12-new-arm-and-x86-cores-are-coming

Jim Keller was very complimentary about the ARMv8 ISA in his talk, saying it has more registers and "a proper three-operand instruction set." He noted that ARMv8 doesn't require the same instruction decoding hardware as an x86 processor, leaving more room to concentrate on performance. Keller even outright said that "the way we built ARM is a little different from x86" because it "has a bigger engine." I take that to mean AMD's ARM-compatible microarchitecture is somewhat wider than its sister, x86-compatible core. We'll have to see how that difference translates into performance in the long run.

http://www.youtube.com/watch?v=eTVnxaXCLg0

see video at 54:50 to 55:20

This is plain false. ARM isn't inherently more efficient. In modern microprocessor designs, for performance requiring markets, that use multiple watts of energy, ISA doesn't matter at all. This is a pure fact -- supported by scientific research. x86 might not be the best ISA of all time, ARM surely isn't either, and x86 is good enough (Intel tried with Itanium, and they failed).

btw Intel's process lead has not helped them in the mobile market where baseband integration is key. The ARMv8 cores are being actively designed into mobile SOCs.

http://www.anandtech.com/show/7925/...bit-socs-with-lte-category-67-support-in-2015

Intel's also catching up quickly in that area.

Qualcomm has standard ARMv8 64 bit CortexA57/A53 cores based top to bottom product stack with integrated baseband in 2015. Qualcomm continues to work on their custom ARMv8 core and will launch in 2016 at 16/14 FINFET.

Competing with Qualcomm is easier said than done. Also you don't realize that the top ARM partners are licensees who design custom cores to differentiate their products - Apple Cyclone, Qualcomm, AMD (K12), Nvidia (Denver), Marvell. That kind of effort is not easy for companies like Mediatek who do not have the resources to design custom cores.

Of course I realize that those companies don't use off-the-shelf ARM IP, but developing your own cores isn't a free lunch.

Intel's node naming is misleading.

http://www.electronicsweekly.com/mannerisms/general/the-intel-nanometre-2013-02/

Intel 22nm has no significant transistor density advantage over the foundries 28nm though it does have performance advantage due to FINFET. Intel's 22nm and TSMC/Samsung/GF 28nm all use single pattern immersion lithography.

Intel's 14nm FINFET and TSMC 16FF+ and Samsung 14 FINFET are all of similar transistor density with a marginal lead to Intel. They use dual pattern immersion litho with a M1 metal pitch of 64 nm. TSMC 16FF+ has a 15% area scaling over TSMC 20nm planar. Samsung 14 FINFET has a similar 15% area scaling over 20nm planar designs.

https://markets.jpmorgan.com/research/email/-kjegkq4/GPS-1336259-0

page 4

http://globalfoundries.com/docs/def...dries-14nm-collaboration---final.pdf?sfvrsn=2

If Intel 22nm FINFET Baytrail is competing with 28nm planar tech based mobile SOCs from Qualcomm then you can figure out for yourself how things are going to be with no process node advantage in 2016.

Lol. Intel's node naming misleading? I advice that you read this very informative article: The Status of Moore's Law: It's Complicated.

I'd call Intel's node naming the least misleading. TSMC, Samsung and GlobalFoundries will all switch from 20nm planar to FinFET and call that transition a new node; from 20nm to 16/14nm. Because, they say, going to FinFET will have similar benefits as a normal node shrink. It hasn't. There isn't any density improvement, except for the updated FinFET Plus that will come later, but I don't call 15% a node shrink.

By this flawed logic, Intel would now have been at a 7nm node, soon 5nm (aka 14nm). Truth is, Moore's Law, which is what those transistor sizes are referring to, doesn't tell anything about performance or power, which is why those foundries call it 14/16nm instead of 20nm FinFET; Dennard scaling does.

So in my opinion, it is just plain wrong of those companies to now completely destroy node names. And why? Because marketing; to fool people (I'd see, people like you, but I don't want to sound offensive). In a few years, they will launch 10nm, and someone who didn't inform himself with articles like those mentioned above, will think they caught up to Intel and its lead, which Intel now has for a long time, has vanished. Unless TSMC proves us wrong, which I strongly doubt, this will not be the case at all. Instead, Intel will even have increased its process lead. But people might not know because the number TSMC has given its node says 10 -- same as Intel.


(In easier words: if you buy a 20nm fab, you upgrade the transistor to FinFET, then buy a new, second fab, which node do you have?)

 

[witeken]

Witeken
all said and done the companies which lost revenue in the last 2 years were x86 PC companies. They were cannibalized by ARM partners selling mobile devices- smartphones, tablets and notebooks(Chromebooks).

The same is now going to extend to servers with ARMv8. You can wait and see how this unfolds for x86 and ARM.

That is something completely different. People obviously want smaller computers; smartphones. So probably less people bought desktops because of that. But not because ARM is better or because desktops now use ARM.

Intel will now take the phone space. But ARM, what will they do? Intel already has lots of solutions for people who want to buy a server, while ARM is still nowhere and has no compelling products that are in any way superior, contrary to Intel smartphone products.

 

[NTMBK]

Intel will now take the phone space.

I still see no reason for that to be true. Poor modem tech, poor GPU tech and a staggering binary translation performance cost are too much for better transistors to overcome- especially when 20nm SoCs are right around the corner, too.

 

[witeken]

Yes, specifically with the Goldmont atom core, the Intel CEO referred to it as having a chassis that allowed easy integration.

How Intel implements this will be very interesting. It would be great to see Intel be able to occupy more niches with atom. (to keep the market competitive and to benefit us as consumers).

With Cortex A57, I can already imagine the multitude of SOCs and niches that could be potentially filled due to the sheer number of partners ARM has (Even Rockchip has licensed Cortex A57).....so Intel with its atom needs to compete.

Brian Krzanich specifically said that Broxton allows them to react extremely quick.

Investor Meeting, from ~20:00.

A lot of the response was: "Well, wait, we didn't think you wanted to be a part of this. Are you here to stay?" So we had to spend some time visiting them and convincing them, yes, we're here to stay.
[...]
We are gonna be more pragmatic, to drive these solution into the market quicker. (..) Think of it as the next-generation Atom. What's different about this, and what we've done to accelerate this, is that this is a core that has a complete new what we call chassis, or basically connectivity that allows to do iterations and derivatives of this core at a very fast clip. Think of it as a core with a common chassis that allows connection of IP both internal and external IP at a very fast rate. The kind of rates that you'd see with external products. The kind of rate our competitors do, this capability. And you've said: why can't Intel. And this product is the kind of product that brings that capability really to reality. Targeted towards mid-15.

 

[cbn]

But ARM, what will they do? Intel already has lots of solutions for people who want to buy a server, while ARM is still nowhere and has no compelling products that are in any way superior, contrary to Intel smartphone products.

We have to look and see where ARM can find niches:

One example I can think of is unRAID NAS. Intel doesn't have a SOC or CPU + PCH with 25 native SATA ports. Sure the experts/IT professionals out there can piece together the necessary SATA cards and port multipliers (what have you) and verify they work properly (after investing a good amount of time). But what about folks who are less tech savvy and want something more appliance-like that can work out of the box with their odd assortment of drives? (Six SATA ports on Avoton is pretty good, but I'm sure there are people who might like 12 native SATA ports or more)

Also we have to consider how the price of hardware affects people in other countries (where labor cost is much lower than hardware cost). $2000 for a server cpu may not be that much in the US where server licensing costs are high and IT professionals are nicely paid....but in other parts of the world the hardware costs will likely make up a higher proportion of the budget.

 

[witeken]

And Apple has shown that they can compete with Baytrail 1 node down. In end 2015 all their competitors will have 14nm/16nm FinFET hybrids producing chips closing on more.

That's like saying that FX-9590 can compete with i7. Yes it might be able to compete, but its TDP is twice as much. Your end-2015 estimate for FinFET seems optimistic to me.

 

[teejee]

Intel will now take the phone space.

Please tell me any sign of Intel taking the phone space.

What high volume phones uses Intel?
What design wins have they taken recently?
Right now Intel has very close to zero market share in phones.

And so far no signs of improvement, i.e. no new design wins.

 

[witeken]

I still see no reason for that to be true. Poor modem tech,

This statement is based on what? As far as I know, this isn't true and Intel's also quickly improving and catching up.

poor GPU tech

Adreno 320-330 performance for Android is more than enough, not really poor or so. Gen 7 is simply the end of its cycle. Cherry Trail will very likely make this statement obsolete, too.

 

[cbn]

Brian Krzanich specifically said that Broxton allows them to react extremely quick.

Investor Meeting, from ~20:00.

A lot of the response was: "Well, wait, we didn't think you wanted to be a part of this. Are you here to stay?" So we had to spend some time visiting them and convincing them, yes, we're here to stay.
[...]
We are gonna be more pragmatic, to drive these solution into the market quicker. (..) Think of it as the next-generation Atom. What's different about this, and what we've done to accelerate this, is that this is a core that has a complete new what we call chassis, or basically connectivity that allows to do iterations and derivatives of this core at a very fast clip. Think of it as a core with a common chassis that allows connection of IP both internal and external IP at a very fast rate. The kind of rates that you'd see with external products. The kind of rate our competitors do, this capability. And you've said: why can't Intel. And this product is the kind of product that brings that capability really to reality. Targeted towards mid-15.

Yes, that is where I got my info from.

Pretty exciting to see what could happen there.

 

[NTMBK]

This statement is based on what? As far as I know, this isn't true and Intel's also quickly improving and catching up.

Adreno 320-330 performance for Android is more than enough, not really poor or so. Gen 7 is simply the end of its cycle. Cherry Trail will very likely make this statement obsolete, too.

Their first integrated modem SoC is still unreleased, and is coming out on the apparently inferior 28nm TSMC node...

 

[witeken]

Please tell me any sign of Intel taking the phone space.

What high volume phones uses Intel?
What design wins have they taken recently?
Right now Intel has very close to zero market share in phones.

And so far no signs of improvement, i.e. no new design wins.

Sure.

The simplest reason that I can come up with is that Intel is simply serious about this space. They're saying all the time that they want to go in that space, and the first thing BK said when he became CEO was that they're going to address the mobile markets. We've now seen Haswell, we've seen Silvermont on Intel's low power 22nm process, we've seen some great products like T100. And we've seen Intel's roadmap: 14nm in 2014 with Broadwell and a completely revamped GPU architecture with 4x as much EUs. A new tock, Goldmont, in mid-2015, only about 20 months after Silvermont (or possibly even earlier if Willow Trail comes in Q1) instead of 24-28, SoFIA. By 2016, Intel will have all products they need, and they will be better than the competition. If they then release a quick follow-up of Broxton and SoFIA on 10nm, their mobile victory is achieved.

And closer in time, we've also seen the first Android tablets with Silvermont and there's a rumor about the Nexus 8, which would be a very great design win.

Intel's the biggest semiconductor company in the world. If they want to be in a market, they're going to be in that market. Design wins aren't the only thing you have to look for to see if they're going to succeed in their mobile campaign, getting design wins is obviously the last step in the process.

 

[witeken]

Their first integrated modem SoC is still unreleased, and is coming out on the apparently inferior 28nm TSMC node...

SoFIA will compete against quadcore A7s, so not having a process advantage is less crucial. Those entry SoCs probably won't get upgraded as fast to 20nm as the high-end.

 

[Arachnotronic]

SoFIA will compete against quadcore A7s, so not having a process advantage is less crucial. Those entry SoCs probably won't get upgraded as fast to 20nm as the high-end.

SoFIA will compete against quad core A53s, not A7s (~50% performance uplift). These entry SoCs will likely shift from 28 LP in 2014 to 28nm HKMG in 2015, so Intel will not have a process disadvantage.

That said, it is a bit premature to assume Intel will most assuredly win mobile. The company has been mostly incompetent at the design of smartphone platforms to date, so until they can prove that they can get a world-class platform out on time, I think a healthy dose of skepticism is warranted.

 

[cbn]

SoFIA will compete against quadcore A7s, so not having a process advantage is less crucial. Those entry SoCs probably won't get upgraded as fast to 20nm as the high-end.

One advantage that I see with SoFIA (compared to cortex A7) is better FPU and single thread.

This should allow a phone with SoFIA to run legacy desktop apps better than the equivalent priced product from ARM.

So in some ways, I am not worried about phone so much for Intel. I think they actually do pretty well on the low end provided the clocks are high enough on the 28nm SoFIA and the linux compatibility is there.

 

[Arachnotronic]

Sure.

The simplest reason that I can come up with is that Intel is simply serious about this space. They're saying all the time that they want to go in that space, and the first thing BK said when he became CEO was that they're going to address the mobile markets. We've now seen Haswell, we've seen Silvermont on Intel's low power 22nm process, we've seen some great products like T100. And we've seen Intel's roadmap: 14nm in 2014 with Broadwell and a completely revamped GPU architecture with 4x as much EUs. A new tock, Goldmont, in mid-2015, only about 20 months after Silvermont (or possibly even earlier if Willow Trail comes in Q1) instead of 24-28, SoFIA. By 2016, Intel will have all products they need, and they will be better than the competition. If they then release a quick follow-up of Broxton and SoFIA on 10nm, their mobile victory is achieved.

Intel's mobile group has given us no reason to think that it can execute that well, so I would maintain a healthy dose of skepticism.

I think Apple's A9 will be a 14/16nm FinFET device, and if this is the case, Intel's process lead will essentially be limited to whatever density advantage it is claiming against TSMC. From TSMC's drive currents presented in its 2013 IEDM paper, 16 FinFET+ should offer equivalent performance to Intel 14nm.

Intel's process lead really is mostly nonexistent when it comes to mobile processors and until I see evidence of such a lead in shipping product, I would again be cautious about statements that Intel's victory is assured.

 

[cbn]

Intel's process lead really is mostly BS when it comes to mobile processors and until I see evidence of such a lead in shipping product, I would again be cautious about statements that Intel's victory is assured.

Ideally Intel could design an entry smartphone chip that would actually be useful even if ARM had the process lead.

I think if they can go after the legacy desktop apps (linux works fine for that) in developing countries there could be some gains.

.....but there is the question of Cortex A53s single thread and FPU. (How well will a converged mobile/desktop linux OS run on Cortex A53 vs SoFIA?)

 

[teejee]

Sure.

The simplest reason that I can come up with is that Intel is simply serious about this space. They're saying all the time that they want to go in that space, and the first thing BK said when he became CEO was that they're going to address the mobile markets. We've now seen Haswell, we've seen Silvermont on Intel's low power 22nm process, we've seen some great products like T100. And we've seen Intel's roadmap: 14nm in 2014 with Broadwell and a completely revamped GPU architecture with 4x as much EUs. A new tock, Goldmont, in mid-2015, only about 20 months after Silvermont (or possibly even earlier if Willow Trail comes in Q1) instead of 24-28, SoFIA. By 2016, Intel will have all products they need, and they will be better than the competition. If they then release a quick follow-up of Broxton and SoFIA on 10nm, their mobile victory is achieved.

And closer in time, we've also seen the first Android tablets with Silvermont and there's a rumor about the Nexus 8, which would be a very great design win.

Intel's the biggest semiconductor company in the world. If they want to be in a market, they're going to be in that market. Design wins aren't the only thing you have to look for to see if they're going to succeed in their mobile campaign, getting design wins is obviously the last step in the process.

sorry, you are just showing me marketing and investor information from Intel. Of course that information looks promising.
Still no signs of any actual progress in phone market. And SOC is probably choosen about a year or so in advance for high volume phones. So why should any sane phone manucturer choose Intel for their 2015 flagship when they have zero track record? It simply doesn't work like that in professional business.

They have so far failed completely to do a compelling phone SOC that any phone company wants to put in a phone.
The SOC is the heart of the phone, and changing from ARM to x86 is a huge step sideway without clear advantages. That's why no phone manufacturers chooses Intel. They prefer the more open ARM system with multiple SOC suppliers.
If Intel wants to suceed in phone market then they should become a foundry and make SOC's for Qualcomm, Apple etc.

 

[witeken]

Intel's mobile group has given us no reason to think that it can execute that well, so I would maintain a healthy dose of skepticism.

Yes, I'm skeptical, but the facts I see make me optimistic. It takes a lot of time to develop products. BK is going all-in on smartphones. As we go further in time, more of his influences, and more of the influences of the companies they acquired and their increased R&D budget, will become clear. SoFIA is one of them, and Broxton is another one. What you call 'incompetent' really is Intel's smartphone team in the 2010 or so time frame.

I think Apple's A9 will be a 14/16nm FinFET device, and if this is the case, Intel's process lead will essentially be limited to whatever density advantage it is claiming against TSMC. From TSMC's drive currents presented in its 2013 IEDM paper, 16 FinFET+ should offer equivalent performance to Intel 14nm.

Why should Intel care about Apple's A9 from an economical point of view? If you want number on Intel's density lead: 10nm vs 20nm FinFET will at least be 2.5x higher density; Intel will be able to put 2.5B transistors in the same area that Apple fills with less than 1B 16nm transistors. But the real price/transistor of 10nm will be (the equivalent of) another node or so lower, because 16nm in fact sees a decrease from 20nm, which is already flat, and those companies pay the foundry tax.

Intel's process lead really is mostly BS when it comes to mobile processors and until I see evidence of such a lead in shipping product, I would again be cautious about statements that Intel's victory is assured.

It isn't BS. You already see it at 22nm vs 28nm. You in fact wrote a whole article about it on Seeking Alpha. If you read my text again, you'll see that I say their victory is assured if Intel releases a fast follow-up of Broxton on 10nm; aka if their investments pay off and their execution is top-notch.

Don't forget that 10nm will introduce exciting new innovations.

 

[Arachnotronic]

What you call 'incompetent' really is Intel's smartphone team in the 2010 or so time frame.

Yep, and now shareholders are paying for it. Otellini's obsessive focus on gross margins is why Merrifield is an unmitigated disaster, and Intel's inability to put a modem on 22nm is why Intel will now be giving TSMC plenty of fab-filling 28nm HKMG business.

Otellini and Perlmutter had no idea what they were up against in smartphones, and now Intel stockholders are paying the price. It's seems that Otellini walked away because he realized "what a mess I've gotten us into" and didn't want to deal with the multi-year clean-up job that Krzanich is going to spend most of his career as Intel CEO performing.

10nm vs 20nm FinFET will at least be 2.5x higher density.

A9 will be a Q3/Q4 2015 event...10nm from Intel may be Q3/Q4 2016.

It isn't BS. You already see it at 22nm vs 28nm.

Really, because I still have yet to see a single 22nm smartphone from Intel on the market.

28nm HKMG has been in the market from TSMC (via Qualcomm in SD800) since May 2013. Intel's best smartphone chip is still a 32nm HKMG mess known as CLT+.

In tablets, the situation is different.

 

[witeken]

sorry, you are just showing me marketing and investor information from Intel. Of course that information looks promising.

This isn't just marketing, it are actual products, or progress. If you want design wins like an SGS5: you won't find those. I didn't want to show you things that don't exist, but actual progress, showing Intel is serious. They have the physics, the roadmap and R&D money on their side. You also can't simply lie to investors.

Still no signs of any actual progress in phone market. And SOC is probably choosen about a year or so in advance for high volume phones. So why should any sane phone manucturer choose Intel for their 2015 flagship when they have zero track record? It simply doesn't work like that in professional business.

No signs of progress? Did you miss MWC? And in one of my earlier posts, I showed a quote of BK assuring the OEMs they're serious about this market. Also a SoCs isn't really chosen a year in advance... for example back in January Samsung didn't even know if it would include a fingerprint scanner, so an SoC probably could still be changed too.

You convince OEMs not by telling stories, but by, like BK said, showing great, leading edge products, and a compelling roadmap (+ they have a track record in the desktop/laptop space with their innovation etc.). Intel has exactly that. If you don't value progress that are not design wins, I can't help you further. But you should know that SoCs aren't made by doing nothing.

In the end, you don't even need a track record. Just a great product, a nice price and as bonus the promise for follow-up products (a roadmap).

They have so far failed completely to do a compelling phone SOC that any phone company wants to put in a phone.
The SOC is the heart of the phone, and changing from ARM to x86 is a huge step sideway without clear advantages. That's why no phone manufacturers chooses Intel. They prefer the more open ARM system with multiple SOC suppliers.
If Intel wants to suceed in phone market then they should become a foundry and make SOC's for Qualcomm, Apple etc.

Intel's working on it. Merrifield is their first serious smartphone SoC. I think the move to x86 from ARM is really exaggerated. I can give you 1 advantage: an OEM can wait until the very last minute to choose between every possible OS. And if they don't like x86, or there's a better ARM alternative, they could easily go back to Qualcomm or MediaTek.

 

[cbn]

On SoFIA, I wonder what kind of clockspeed Intel will get out of the Silvermont cores?

Do we know what TSMC 28nm process it will use and what the Target TDP will be?

(I am hoping it is a decent enough clockspeed so it could be used as a desktop as well.)

 

[witeken]

Yep, and now shareholders are paying for it. Otellini's obsessive focus on gross margins is why Merrifield is an unmitigated POS, and Intel's inability to put a modem on 22nm is why Intel will now be giving TSMC plenty of fab-filling 28nm HKMG business.

Otellini and Perlmutter had no idea what they were up against in smartphones, and now Intel stockholders are paying the price. It's clear that Otellini walked away because he realized "what a mess I've gotten us into" and didn't want to deal with the multi-year clean-up job that Krzanich is going to spend most of his career as Intel CEO performing.

This is overblown. BK will have a full line-up of products from IoT to tablets by the end of 2015, a mere 2 years after he became CEO. PO did the foundation of the work for smartphones.

A9 will be a Q3/Q4 2015 event...10nm from Intel may be Q3/Q4 2016.

Again, what does Intel care about A9? If we take the 64-bit adaption delta between Apple and Android as an indication, then the first 20nm FinFET products that will compete against Intel will come out at the same time as 10nm. That might not be the case, but saying that Apple will release FinFETs in Q3 15 and Intel 10nm in Q4 16 is very optimistic + pessimistic.

Really, because I still have yet to see a single 22nm smartphone from Intel on the market.

28nm HKMG has been in the market from TSMC (via Qualcomm in SD800) since May 2013. Intel's best smartphone chip is still a 32nm HKMG mess known as CLT+.

In tablets, the situation is different.

The recently announced Merrfiefield. It's like the people who are now also claiming, at the end of its cycle, that Gen7 is very bad.

 

[cbn]

One more question:

On the iGPU side of things for future SoFIA-class smartphone chips, what is the lowest number of EUs Intel could use for Gen graphics ? Is it two EUs?

 

[Arachnotronic]

You also can't simply lie to investors.

Intel -- like all public companies -- sometimes makes "technically true" statements that don't mean what they might appear to.

Also a SoCs isn't really chosen a year in advance... for example back in January Samsung didn't even know if it would include a fingerprint scanner, so an SoC probably could still be changed too.

Sure it is. According to Broadcom, the selection process for 2015 handset SoCs is beginning now.

No signs of progress? Did you miss MWC?

Yeah, Intel paper launched two SoCs that wouldn't show up in devices until well after the launch.. Merrifield is an unmitigated failure technically and commercially, and Moorefield still lacks imaging performance and integration that competing Qualcomm chips have. In this case, Intel's magical process lead amounted to nothing, and hordes of investors who bought Intel's spin line about how great its manufacturing is and what it would do in Silvermont and beyond got tricked because what Intel didn't mention until recently is that they didn't port any of the critical smartphone IP to 22nm.

Intel's working on it. Merrifield is their first serious smartphone SoC. I think the move to x86 from ARM is really exaggerated. I can give you 1 advantage: an OEM can wait until the very last minute to choose between every possible OS. And if they don't like x86, or there's a better ARM alternative, they could easily go back to Qualcomm or MediaTek.

First serious SoC? Ah, I guess Medfield and Clover Trail+ were just warm-up exercises. Is that why Intel projected a large revenue increase in 2013 over 2012 in mobile (that ultimately didn't materialize)?

The Android/Windows support is a canned Intel PR argument that I think isn't as important as one would think. Windows Phone doesn't even run on X86, which is yet another big disadvantage for Intel. And Windows tablets? Maybe...but I wouldn't hold my breath on any OEM caring that they can use the same hardware for a Windows and an Android flavor of a tablet.

Look, I want to see Intel succeed, but let's call it as it is...Intel has a lot to prove in mobile.

 

[monstercameron]

didn't take long for this thread to be all about Intel...sigh.