TSMC and ARM Tape-Out First ARM Cortex-A57 Processor on 16 nm FinFET Technology

[krumme]

Well if the new atom is fast and the phones are not pricier than comparable ARM based phones I would sure prefer it because I would mainly be browsing on such a phone and better single threaded performance is obviously very important for that. RAZR I is priced well IMHO, not more expensive than similar phones.

I agree about the GPU part. However AFAIK all ported games also run on crappy gpu of RAZR I so no big issue there except lower quality settings.

Yes - you an i will buy this kind of phone. But how big is that segment? And why should Samsung give Intel control over their portfolio. Its simply not a strategic possibility.

What kind of profit do you thin Samsung is going to have q2? Do they want that to be threatened?

Add to that that a57 is comming to market at 20nm just months after the new atom or at a cheap 28nm. It means atom arrives at the worst possible time.

 

[CHADBOGA]

Add to that that a57 is comming to market at 20nm just months after the new atom or at a cheap 28nm. It means atom arrives at the worst possible time.

LOL What nonsense.

 

[Homeles]

Add to that that a57 is comming to market at 20nm just months after the new atom or at a cheap 28nm. It means atom arrives at the worst possible time.

Are you nuts? I don't know what kind of planet you live on, but here on Earth, it is better to have the best you've got out before your competition does. 22nm Silvermont should do very, very well against a 20nm A57. Then when 14nm Atom comes out about 6 months later, ARM's dominance in the mobile space will begin to crumble.

 

[mrmt]

I dont think its relevant to discuss arm servers before we see what a57 is capable of compared to the new Atom. And even at that time ofcource its of minor importance. I agree with you. My post was just a reaction to the typical answer completely ignoring the history even for Intel. Even the server market reach maturty at some time, i just cant see it comming the next 5 or even 10 years.

The fact that Intel is moving Atom to its prime process and bringing Core to power consumption levels is a testament on how serious Intel is taking the ARM threat, and they aren't stopped.

Sure, they could be moving faster, but when you have the kind of elephantine corporate structure like Intel do, it takes a while to turn around the ship. Just look how long it took to Intel to redesign Atom. It's been 4 years since Intel promised us the redesigned 22nm Atom. But OTOH, once the wheel is spinning, Intel is a huge force. As of now I prefer to wait and see. I'd like to see whether Intel will field a product that can match ARM on its own turf.

As for the opposite, whether ARM will develop products that can compete in Intel's turf, I prefer to wait and see too. I just don't buy a lot of the A7 claims, as I don't buy a lot of the ARM hype. If A7 is 80% of A15 performance at a third of the costs, why bother developing A15 in the first place? Something doesn't add up here, and the fact that people like Hans de Vries and some other AMD arch-fanboys are hyping the thing just makes it looks more suspicious.

For me, A15 was a deception. I was expecting Atom performance in a more power efficient fashion, and I got none. By looking at the ARM ecosystem, we can see that costs and margins are starting to bite already. Except for Qualcomm, everyone dropped shipments. People forget that as soon as IC grow in complexity the costs also grow in an exponential fashion, and Krait shouldn't be a $10 SOC as the rest of the ARM appliance chips.

To better illustrate what I'm talking about, take microservers. If you want a microserver now, your best chance is with Intel Xeon and Atom, not with ARM. Yet people see this as an inherently ARM turf that Intel would have no chance.

And A57? What are the trade offs between the higher performance and their development costs, and design/validation overhead, and die size, etc? Nobody answered this question yet, and until we see in 2015 actual product, I see wait and see as a very prudent approach.

I guess we are more or less in the same tune here, but just different perspectives. You see that ARM is growing faster than it should and effectively threatening Intel, while I see ARM movements as not vitally threatening, and that Intel is fielding chips in the ARM turf that are up to the task of open a beach head on the ARM market.

Just don't expect Intel to be steamrolled by ARM just because they aren't hyping their Atom chips the way ARM does with A57. It's just that Intel has only Intel to worry about in their product line, there is no Qualcomm Atom. But in the multitude of ARM manufacturers out there, if the small ones don't advertise themselves, who will notice them?

 

[mrmt]

http://seekingalpha.com/currents/post/931051?source=email_rt_mc_readmore

H-P (HPQ -0.9%) unveils its first Moonshot system, the fruits of a much-hyped effort to develop the kind of densely-packed, low-power systems many Web/cloud service providers have embraced. The first Moonshot server uses Intel's (INTC) S1200 (Centerton) Atom CPU. 45 of them fit into a small enclosure containing shared power/networking/storage components, and H-P claims 1,800 servers can fit into a rack. AMD and ARM-based (ARMH) servers will arrive later. The likes of Google, Facebook, and Amazon probably won't adopt Moonshot, but some enterprises and service providers hoping to emulate those firms might.

===========

See?

 

[Exophase]

I've never seen anyone say A7 is 80% of A15 performance or anything remotely similar. I've seen claims that A7 offers around 70-80% of IPC that Cortex-A9 does (Hans made one himself) and this is pretty easily defensible.

A15 on the other hand has on average offers much higher IPC than A9. That is only perf/MHz, A15 designs so far have been clocking a little higher than the highest clocked A9s and much higher than the highest clocked A7s. The design itself has a lot more potential for high clocks.

The reason why A7-only SoCs can be "good enough" for some is because they're offered in markets that were much further behind fairly recently. They're coming from SoC vendors like MediaTek, Allwinner, Amlogic, and so on based in China and Taiwan. And they're easily competitive with Intel's 1GHz Lexington SoC that is also meant to address the lower end.

I don't think A15 was a deception at all. Those of us following the uarch long before it was released expected that it'd be a compromise trading higher peak perf for lower perf/W (at least in higher clocks), when compared to Cortex-A9 on the same process. Looking at what we knew about the designs this was just common sense. For instance it's expressed in this article: http://www.beyond3d.com/content/articles/111/ Notice that Arun was also eagerly anticipating a heterogeneous multiprocessing solution from ARM - this was long before big.LITTLE was announced and ARM hadn't given any kind of hints that it was pending so that shows that at least some of us thought it was a good idea purely on its own merits and not merely because it was what ARM was doing (back then we were actually wondering if pairing Cortex-A15s with Cortex-A5s made sense, but it hit some problems that the mix of Cortex-A7 solves).

As for "Atom performance", A15 has easily exceeded that in competing products. What we're currently lacking is a good model of the entire perf/W curve. See how much more power Cortex-A15 uses when its clocks are limited to give performance equivalent to some comparable Atom platform (Medfield, Clovertrail(+), etc) at its peak clocks. That'll give a much more balanced picture.

 

[krumme]

LOL What nonsense.

[redacted]

The next time you threaten another member may very well be the last time you ever post here. We will not tolerate that kind of crap.
-ViRGE

 

[krumme]

Are you nuts? I don't know what kind of planet you live on, but here on Earth, it is better to have the best you've got out before your competition does. 22nm Silvermont should do very, very well against a 20nm A57. Then when 14nm Atom comes out about 6 months later, ARM's dominance in the mobile space will begin to crumble.

Well excuse me earth professor but i didnt notice atom arived before the competition.
What is it now the names of the nerd phones that carries this technology.
RAZRRGR
RAZZGR
BZGARRAZ
GRAAAAAAAAZZZ 1+
?
If this single threadded monster that samsung and apple desparately need, is so fast with its brilliant eu graphics, as your crystal ball predicts surely TI and the likes will feel the heat.

 

[Lonbjerg]

[redacted]

Reported...

 

[R0H1T]

Maybe some of you guys really need to stop talking out of your backside ~

Samsung Octa
MediaTek MT6589

If you extrapolate those numbers wrt power consumption, the Cortex-A7 wouldn't look half as bad as what you're claiming & just for the record anyone who claims that this forum isn't biased towards Intel is clearly lying through their teeth !

 

[krumme]

Maybe some of you guys really need to stop talking out of your backside ~

Samsung Octa
MediaTek MT6589

If you extrapolate those numbers wrt power consumption, the Cortex-A7 wouldn't look half as bad as what you're claiming & just for the record anyone who claims that this forum isn't biased towards Intel is clearly lying through their teeth !

Thanx for link.
The MediaTek MT6589 is the first A7 quad core on the market, made on 28nm lp with all integrated. Surely in this Mediatek a dirt cheap implementation sucking power and perform to the lowest side, and even then performs quite respectfully. This arch performs without advanced powergating and granularity, making production far simpler and reduces time to market. And it doesnt in any way need 20/16nm.

Within a year all low and midrange will run on this arch, replacing single and dual core A5. Its highly power efficient and each cpu core with L1 is 0.45mm2. Even for performance judging by the numbers, in many situations, a step up from dualcore A9.

Its the gate for good experience on 4.1 and 5.0 for all Android devices and makes quadcore the norm. Absolutely the most influential cpu for the mass market this year.

 

[Medu]

As for the opposite, whether ARM will develop products that can compete in Intel's turf, I prefer to wait and see too. I just don't buy a lot of the A7 claims, as I don't buy a lot of the ARM hype. If A7 is 80% of A15 performance at a third of the costs, why bother developing A15 in the first place? Something doesn't add up here, and the fact that people like Hans de Vries and some other AMD arch-fanboys are hyping the thing just makes it looks more suspicious.

Intel NEEDS to compete with ARM's low power products as they are eating into Intel's historical user-base. OTOH ARM, and it's partners don't need to compete with Intel 'Core' family as it's not a treat to their business model.

A7 is only 1/2 as fast as A15, for 1/3 of the power so what doesn't add up is your info.

22nm Silvermont should do very, very well against a 20nm A57. Then when 14nm Atom comes out about 6 months later, ARM's dominance in the mobile space will begin to crumble.

Who is going to use it?
Apple? They design their own chips.
Samsung? Also design/manufacture their own chips.

Currently that is all that matters in the mobile space. Chinese companies are getting more prominent but I don't see them switching to a more expensive Intel solution.
Even if/when Intel get a very good mobile solution it is going to be very hard to get any of the major players to switch to it.

 

[mrmt]

Who is going to use it?
Apple? They design their own chips.
Samsung? Also design/manufacture their own chips.

Not Apple, but Samsung? Probably. Despite having in-door solutions, Samsung always reach Qualcomm for some products. If they chose Qualcomm, why can't they go Intel?

And there's always the second-tier Android manufacturers that once added up are a significant part of the market. LG, Sony, ZTE, etc.

 

[lagokc]

Not Apple, but Samsung? Probably. Despite having in-door solutions, Samsung always reach Qualcomm for some products. If they chose Qualcomm, why can't they go Intel?

And there's always the second-tier Android manufacturers that once added up are a significant part of the market. LG, Sony, ZTE, etc.

Samsung has always been very pragmatic. While Apple might stubbornly use their inhouse designs whether they're superior or not, Samsung will switch to Intel chips in an instant if it would give their final product a competitive edge.

 

[R0H1T]

Samsung has always been very pragmatic. While Apple might stubbornly use their inhouse designs whether they're superior or not, Samsung will switch to Intel chips in an instant if it would give their final product a competitive edge.

Hardly, they'll go for higher margins especially if the performance difference is only a few percent higher as compared to the top ARM chip. What others haven't taken into account is that the x86 port of Android isn't so much popular as the ARM one & then you'll have other problems with proprietary software on such devices, this is one of the main reasons why winRT didn't take off !

 

[krumme]

Not Apple, but Samsung? Probably. Despite having in-door solutions, Samsung always reach Qualcomm for some products. If they chose Qualcomm, why can't they go Intel?

And there's always the second-tier Android manufacturers that once added up are a significant part of the market. LG, Sony, ZTE, etc.

It just doesnt add up only selling to LG and Sony high end models. Its nothing.

Graphics is the driver in the mobile market now, and i see no signs of that changing. If the 2014 variant use the same ach as usual, there is simply no market motive to use Intels solution.

Intel needs Apple on their team as Apple is the company that can push new technology and demands. What do i know, some apple exclusive aps, only running on Intel because of the - hopefully - very fast hardware? - what do i know, it looks absolutely hopeless, as Apple want to control as much as possible themselves..., and are not dependent on the sort of performance selling Intel can do on their traditional markets.

Intel needs something to push their tech, and they need to create demand. Because its simply not there as it is now. No on the consumer side, and then add the OEM is one big wall.

Yes Samsung will probably make an Intel phone with some unknown OS, because they just about cover even the smallest niches on the market, but thats not going to make a difference. Besides Intel is scared to death about Samsung.

 

[R0H1T]

It just doesnt add up only selling to LG and Sony high end models. Its nothing.

Graphics is the driver in the mobile market now, and i see no signs of that changing. If the 2014 variant use the same ach as usual, there is simply no market motive to use Intels solution.

Intel needs Apple on their team as Apple is the company that can push new technology and demands. What do i know, some apple exclusive aps, only running on Intel because of the - hopefully - very fast hardware? - what do i know, it looks absolutely hopeless, as Apple want to control as much as possible themselves..., and are not dependent on the sort of performance selling Intel can do on their traditional markets.

Intel needs something to push their tech, and they need to create demand. Because its simply not there as it is now. No on the consumer side, and then add the OEM is one big wall.

Yes Samsung will probably make an Intel phone with some unknown OS, because they just about cover even the smallest niches on the market, but thats not going to make a difference. Besides Intel is scared to death about Samsung.

I get where you're coming from, Intel needs someone like MS(they basically just put them on the desktop market for three decades now) on the mobile front to counter the strong headwinds they face from ARM et al. I respect where they stand in the semiconductor industry & I'll always pick the most power efficient product(Intel/ARM/AMD/Nvidia) within my budget but mobiles/tablets are virtually commodity products now & the cheapest solution there almost always wins !

 

[mrmt]

It just doesnt add up only selling to LG and Sony high end models. Its nothing.

Not high end, upper end. And not only Sony or LG, but LG + Sony + ZTE + Motorola... this can make a significant number.

Graphics is the driver in the mobile market now, and i see no signs of that changing. If the 2014 variant use the same ach as usual, there is simply no market motive to use Intels solution.

NVidia results clearly disagree with this statement.

 

[krumme]

Not high end, upper end. And not only Sony or LG, but LG + Sony + ZTE + Motorola... this can make a significant number.

That assumes a57 on nm is not taking a huge toll. And again for that to happen, Intel needs to lower prices, and then i dont know what interest they have to be there, at google play or IP app store driving business.

Only the high-end can pay for anything remotely interesting, because the money is going everywhere but the cpu also on a high-end phone, all the soc is gpu or to come for the fat high-speed soc - LTE/3g+.

On the other hand if they dont succed, they dont have much to lose as the capacity can be used for servers and desktop. Thats probably the main reason to go mobile.

NVidia results clearly disagree with this statement.

No. Making a quad core on 40nm was a huge marketing win. I dont know how that helps Intel. Go convince consumers an s4 pro was better than tegra3. There will not be anything less than quad on the future mobile market except niches for nerd phones. Besides it might be that NV sells (edit: sold) at mobile now, but how is that going to help them? - its not like there is profit.

 

[lagokc]

Hardly, they'll go for higher margins especially if the performance difference is only a few percent higher as compared to the top ARM chip. What others haven't taken into account is that the x86 port of Android isn't so much popular as the ARM one & then you'll have other problems with proprietary software on such devices, this is one of the main reasons why winRT didn't take off !

x86 isn't as popular as ARM on Android because there is only one chip supplier and Intel's Atoms are an antiquated design on an antiquated process at this point. Once Intel releases the modern OoO redesigned Atom on the 22nm process there is a good chance they will be more than just a few percent faster than the competing ARM chips. The one nice thing about Android using Java for pretty much all of its applications means none of them have to be rewritten to switch from ARM to x86 (granted it means everything runs a bit more slowly).

 

[Exophase]

I don't think Samsung is chasing CPU performance at all costs at all, given Exynos 5 Octa Galaxy S4s have higher CPU performance than the Snapdragon S4s. Their main motivation here is more likely price, so Intel is going to need to be very competitive on this front to get Samsung's attention. The big price advantage in Qualcomm's SoCs is due to baseband integration. Intel will need at least that same level of integration to be appealing. They're getting there but I don't know how long it'll take, and I don't know what Samsung is working on in-house (if anything).

The other issue is fab capacity; the rumors are that Exynos is supply constrained. But once Samsung loses Apple wafers they'll have a huge gap to fill and Exynos is the most rational choice.

NVidia results clearly disagree with this statement.

How do you figure? nVidia's strong heritage in high end discrete GPUs hasn't necessarily translated into a big advantage in low power SoCs. Tegra 3's GPU performance wasn't class leading when it came out, and the situation has gotten much worse since then when other vendors have updated multiple times where nVidia hasn't. They're now pretty much last place:

http://www.anandtech.com/show/6877/the-great-equalizer-part-3/2

The big advantage nVidia had was better drivers. Much like the big compiler advantage Intel had a few years ago (in GCC, to be specific) this has been rapidly diminishing.

Tegra 4 looks to be a huge move forward, but that's not out yet. And it still has some compromises, lacking some features that everyone else has or will have soon, like any kind of OpenCL support.

 

[sontin]

How do you figure? nVidia's strong heritage in high end discrete GPUs hasn't necessarily translated into a big advantage in low power SoCs.
Tegra 3's GPU performance wasn't class leading when it came out,

Sure it has. Tegra 2 was class leader with Samsung and SGX540 and Tegra 3 has the fastest Android GPU for nearly 6 months. And only then there was only Samsung and Mali faster. Snapdragon needed 12 months to be faster. Krait v1 is on par with the ULP Geforce in Tegra 3.

and the situation has gotten much worse since then when other vendors have updated multiple times where nVidia hasn't. They're now pretty much last place:

http://www.anandtech.com/show/6877/the-great-equalizer-part-3/2

If you look at that: The iPad mini which came out 6 months ago is not really faster than Tegra 3.

Tegra 4 looks to be a huge move forward, but that's not out yet. And it still has some compromises, lacking some features that everyone else has or will have soon, like any kind of OpenCL support.

With nVidia's Dev support Tegra 4 will be the plattform for gamer in the future. There are nearly 20 Tegra 4 games in development which will bring a huge amount of hype to the SoC.
People don't care about OpenCL, OpenGL ES 3.0 or >24bit fp precision when there are no games out which showing great graphics.

 

[Exophase]

Sure it has. Tegra 2 was class leader with Samsung and SGX540 and Tegra 3 has the fastest Android GPU for nearly 6 months. And only then there was only Samsung and Mali faster. Snapdragon needed 12 months to be faster. Krait v1 is on par with the ULP Geforce in Tegra 3.

When I say class I meant tablets in general, not just Android tablets, so was considering Apple A5.

If you look at that: The iPad mini which came out 6 months ago is not really faster than Tegra 3.

GPU certainly is, and why should you compare a lower end part from Apple vs nVidia's best offering?

With nVidia's Dev support Tegra 4 will be the plattform for gamer in the future. There are nearly 20 Tegra 4 games in development which will bring a huge amount of hype to the SoC.
People don't care about OpenCL, OpenGL ES 3.0 or >24bit fp precision when there are no games out which showing great graphics.

nVidia would certainly like to make you think so but I personally doubt it'll be dominating gaming in tablets and phones, even with nVidia spending TWIMTBP money on it.. People will care about OpenCL when it's accelerating non-GPU tasks, and it's not like Tegra 4 is in a world apart in raw capability (despite nVidia's questionable slides on the matter) Rogue in particular is going to be very competitive on all fronts and IMG has been eating up more and more market share outside of Apple.

 

[aaksheytalwar]

Is Atom Intels only product for cell phones in 2013?

 

[sontin]

When I say class I meant tablets in general, not just Android tablets, so was considering Apple A5.

And A5 has only 1/4 of the CPU performance for 30% more GPU performance. And nVidia is not in competition with Apple. Even if Tegra 3 would 2x faster certain people would not buy an Android product.

GPU certainly is, and why should you compare a lower end part from Apple vs nVidia's best offering?

In GL 2.7 Nexus 7 and Mini on par with each other.

nVidia would certainly like to make you think so but I personally doubt it'll be dominating gaming in tablets and phones, even with nVidia spending TWIMTBP money on it..

It will easily. If people not interested in a Samsung or Apple device they will look out for Tegra 4 if they playing games.

People will care about OpenCL when it's accelerating non-GPU tasks, and it's not like Tegra 4 is in a world apart in raw capability (despite nVidia's questionable slides on the matter) Rogue in particular is going to be very competitive on all fronts and IMG has been eating up more and more market share outside of Apple.

Rouge comes next year at the same time like Logan. And winning market share? Samsung E5 will only be in a few products and has less performance than Tegra 4. Qualcomm dominates the Android smartphone and nVidia the Android tablet market. There is no room for IMG anymore.

Sorry for offtopic but i thought it was necessary to correct certain points.