Deep inside Intel's new ARM killer: Silvermont Intel has released details about its new Silvermont Atom processor microarchitecture, and — on paper, at least – it appears that Chipzilla has a mobile market winner on its hands. Yes, yes, we know: you've heard it all before, from Menlow to Moorestown to Medfield. Intel has made promise after promise that its next Atom-based …
FFS, am I the only one who tires of the bovine excrement that is "5x improvement in power savings". If they mean it uses 1/5 the power then say so. Other reports I've seen say it uses 5 times less power which would mean that it actually produces 4 times the power that the processor in comparison uses.
I currently have a little Acer netbook that has an Intel Atom N455 "Pineview" microprocessor in it, and I love it. Understandably, the little netbook is no processing powerhouse, but then again I don't need it to be. Instead, I like its light weight, small size, portability, and adequately-sized keyboard. It's great for taking notes at developer user-group meetings, it's satisfactory for browsing the web, and when I am working out of a hotel room I can easily type e-mails, short documents, or edit Excel spreadsheets on it-- all without having to lug around a much larger and heavier laptop, and without having to payout hundreds upon hundreds of dollars to buy an expensive "Ultrabook."
Why do I bring this up? Well, I just can't help but think about what could have been. With how much I like my trusty little netbook now, I would absolutely *love* to buy one sporting one of these new "Silvermont" Atoms in it. Think about it-- if they could produce a netbook with more than 3-times the performance of the current ones and still somehow keep its cost within the typical netbook range of $250, that would be a real winner as far as I'm concerned. Unfortunately, that will almost certainly never happen. To the best of my knowledge all of the current big PC manufacturers have completely discontinued their netbook lines, and the netbook form-factor has been eagerly declared as "dead" again and again and again by the media. To think that I would have to shell out some serious $$$ to buy an overpriced Intel-based tablet or "Ultrabook" to experience the benefits of these new chips in a portable form factor is just depressing.
I can only agree and disagree at the same time. I too had a netbook, although I jumped aboard teh ship early and got an asus 901, one of the earlier mainstream netbooks. (or was it 701, can't remember) I got it mostly for the low price and the fact I could take notes on it at uni.
It was a nice device to be sure, the only problem for me was still the speed. I would type faster than it would pick up my typing and honestly I couldn't get used to that teeny tiny keyboard. The best way to match one of those good old netbooks these days is to buy something like a transformer, but as you've stated these transforming tablets are incredibly overpriced when you consider it's just a tablet with a cheap keyboard attached most of the time.
I'd actually hope to see a few more combo laplets come out at a lower price point in the future. But that's just me.
I've just got myself one of the ARM-powered Samsung Chromebooks for £230. So far I'm favourably impressed by the keyboard and by the performance (no problem with rendering complex web pages or playing Adobe Flash games). I'm not sure I like the limitations imposed by the software; I might install Ubuntu instead. And I've not tested the battery life. Still, the XE303C12 is perhaps the nearest you can get today to the netbook that might have been.
Not really, or at least not for iOS. The default option in XCode is to build your iOS app for x86 and run it in the iPhone Simulator software in OS X. I'd wager this is how most iOS app testing is done.
Building for different architectures is only a big deal if you're on Windows.
The joke here is that Intel preached for years that the x86 architecture matters, as it is backwards compatible to existing software without any emulation/recompilation.
Given that the big majority of the mobile market software is compiled for ARM architecture, Intel has to compete on performance and price, rather than on exploiting a locked in user base.
Ironically Intel helped the transition to ARM, by delivering a rather good performing X-Scale ARM based processor product line, not so long ago. Before there were competing architectures like MIPS in the mobile market.
So yes, most software can be easily recompiled to any architecture nowadays. ARM or X86 doesn't really matter.
Coat - because it is time to leave after explaining a joke...
Atom has always had a hefty margin to maintain Intels bottom line, and while this new architecture looks impressive on paper I seriously doubt it will be competing at the same price points as ARM SoCs unless Intel are willing to take the hit.
Also, AMD were publicly discussing tri-gate transistor tech a few years before Intel, it will be interesting to see what they can achieve with it both in x86 and ARM, plus others such as TSMC are already up to speed with tri-gate so I doubt ARM will lose much sleep over Silvermont.
Atom has always had a hefty margin to maintain Intels bottom line
I don't think that's true. I think Atom has considerably lower margins than the "Core" chips and the "Xeons" which is one of the reasons why increased sales depressed profits last quarter. Atom's have been kept castrated to stop them eating the higher margin business. If the new chips are getting the same features as the more expensive ones then we can hope for significantly cheaper ($50 - $100) Atom-based notebooks but the chips will still be too expensive for phones.
The margins are higher relative to the margins on ARM SoCs. This isn't such an issue in higher value non-mobile applications, but for mobile, where Atom tries and fails to compete more directly with ARM and the volumes are so much higher, unless Intel cut their margins to suit their cloth they will continue to struggle against ARM. Mobile devices are already "fast" enough, so does Silvermont really bring enough to the table to justify a higher BOM price than an equivalent ARM SoC?
Price is key here.
Apple and Samsung et al vertically integrate to extract the profit from all bits of the vertical stack. Which of these megacorps wants to give Intel a slice of profit?
The code for phones has to be rewritten because the gui form factor is different. So there's little portability to be gained there - not much incentive for devs to pick it up.
Windows tablets could be a target market, with RT dying off, and x86 W8 going everywhere. Low-end CPUs are a problem for MS but perhaps they could license by CPU speed. Their problem will be the loss of business desktop revenue if things move to VDI where MS are currently double-dipping on desktops.
The other biggie is low-power stuff. The home server would be a great market, running x86 on low power systems, though I doubt its cheap enough to beat the Pi taped to the back of the telly as an access device.
Moonshot-type devices are also an option but massively parallel cpus are a significant architecture change, Windows is probably not an option without a major licensing change and unit price counts more when you're scaling up. With such devices being custom-made, I can't see HP deciding to give Intel more cash for atoms. Atom would have to be not just better, but an order of magnitude better.
I still want to see a home-based blade system with a low power server blade and a high-power desktop blade in the same box. I want to be able to power off my i7 and GPU card without affecting my always-on media and sync server, while both are on, the PCIe 3.0 running as a network between them would be great.
So what would an ARM processor fabbed at 22nm and using tri-gate tech do then?
Intel cannot win this one if all they can do is throw bleedin' edge design tricks at the problem to break even, as all the opposition has to do to beat them is play catchup when the tricks used go mainstream. Face it Intel, x86 has just too much bloat and baggage for the low-power, small form factor market.
I was wondering the same - what is the performance comparison of ARM vs x86 on the same fab technology?
Also, as others have pointed out, ARM is generally a lot cheaper to buy/licence than Intel's offerings, so this had better be really good to make it.
And what about SoC builds, will Intel be offering to integrate customer's hardware on the same fab as this new CPU?
x86 instruction translation is not a significant factor and hasn't been for years. Intel started from a position of performance at all costs whereas ARM started at the opposite end of the power/performance spectrum. We are finally reaching the point of crossover where A15 cores can be as power hungry (and performant) as x86, and x86 can play in the low power realm.
Serious competition should be good for everyone.
"A15 cores can be as power hungry (and performant) as x86"
I'm willing to believe this, for some reasonable values of x86.
I'd be interested in seeing comparative code density figures - how much memory does application X take on x86 vs the same code on a suitably chosen ARM (and suitably chosen compiler, optionally taking into consideration any essential runtime support such as a Linux kernel, Busybox, etc).
Code density is important in some cost-sensitive and power-sensitive market sectors, markets which Intel have failed to address so far, markets which are massive in terms of units shipped. In some of those sectors, the more functionality you can fit into a given amount of memory, the better. And the less memory your code needs, the less power you can get away with, and the cheaper the system might be, AND the better it may perform (because of cache and main memory bandwidth effects). And you can save on battery cost/weight or get longer battery life, and and and.
ARM have had years of optimising code density (in comparison with classic RISC). I don't really know enough x86 (yet) to be able to comment on this aspect. Any volunteers?
"x86 can play in the low power realm"
Really? Other than a handful of "Intel Inside" mobile phones, where is the *evidence* for this? Evidence means something objective and preferably easily repeatable, not a few fact-free "articles" from an advertising-supported vendor-dependent site (mentioning no names here).
Then as others have pointed out already, there's a world of difference between what Intel calls an SoC and what the ARM world calls an SoC.
E.g. has anybody done a teardown yet on an x86 mobile phone and compared it with what's in an ARM mobile phone aimed at the same market (or price point)?
"will Intel be offering to integrate customer's hardware on the same fab as this new CPU?"
Wasn't that answered in the press relarticle?
No, I don't suppose it was (was there much point reading pages 2 and 3?).
Here's the Intel original press release; it also includes a link to the webcast:
http://newsroom.intel.com/community/intel_newsroom/blog/2013/05/06/intel-launches-low-power-high-performance-silvermont-microarchitecture
Anyway, without the option of much more integration than Intel have previously offered in their own designs [1], they still don't stand much chance on merit alone. I'm sure they'll be able to strongarm a couple of x86-dependent designer/manufacturers into doing extended prototype runs, just as Intel did with their previous generation of smartphone chips. Meanwhile, where there is still free choice, the rest of the market looks elsewhere.
[1] The ARM-based Intel-badged IXP422 etc Intel products weren't bad for their time (2005ish) but Intel sold them off along with most of the rest of their ARM business.
Usually, such claims are based on maximum theoretical performance -- driving all processing units at full load. You may be able to get very close to that with specially-designed benchmark programs, but it is not realistic to get anywhere near this in "real" code.
My guess is that the actual performance gain is around x2 for graphics-intensive tasks and x1.5 for tasks that mostly involve the CPU alone. Not a bad gain, but not x3 overall.
Is your guess based on anything other than past experience of what was promised versus what was delivered? And if so, can you actually back that up with a direct comparison of previous promises versus what was actually delivered?
If you're going to discount their numbers on that basis, then they can only respond by increasing the promise! If they had promised 2x performance, you would have discounted it to 1.5x. If you're just going to pull numbers out of the air, why pick 2x? Why not 1.5x or 1.2x?
Well, I can't say if his numbers are right - but I agree with the feeling.
You see, it's marketing. You (marketing guy) are paid to make the product look good. How do You do it?
1) Take an average measure of your opponent.
2) Craft a test where Your product excels
3) Compare them.
It's not just Intel that does it. All business are guilty of this kind of thing. If it was reverse - and were a press release from ARM - I would think the same thing. Just reversed.
This post has been deleted by its author
I imagine that for an ARM processor to get the benefit of tri-gate technology it would require a re-design from a very low level so a future generation of ARM processors may use it but it won't be the current processors re-engineered.
Even so if I where a hardware manufacturer I wouldn't be jumping onto the Intel bandwagon just yet. Maybe if it turns out to open a real world advantage that sticking with ARM will be not able to close within a reasonable time frame then maybe. But then again, given Intel's track record with customer focus maybe not.
so a future generation of ARM processors may use it but it won't be the current processors re-engineered.
Correct on the first count, wrong on the second... :)
Samsung completes tapeout of Cortex-A7 using 14nm FinFET process technology (Feb 2013)
TSMC tapes out Cortex A-57 using 16nm FinFET process technology (April 2013)
I'd still be concerned about the graphics capability. ARM's killer feature is that you can license whatever core you want, pair it to whatever GPU you want and then farm it out to the lowest bidding fab. Intel have come a long way with their embedded graphics, but the GPUs theyre shipping with Haswell aren't exactly power-efficient when it comes to the mobile world.
Also, even if all their boasts about 3-5x improved performance turn out to be true, that's still a little on the light side to run full-fat Windows and x86 applications.
What is the power consumption of a system using these chips and how does it compare to a similar ARM setup?
From what I can recall whenever Intel have gone on about reduced power consumption and performance per watt compared to ARM, they have compared a single CPU, conveniently leaving out the rest of the chipset, to an ARM system and not done that well either. Are Intel doing the same thing again?
The Intel-based Motorola has similar power/performance figures to ARM-based phones. It has better single-threaded performance, worse graphics and worse task switching but would feel as good to a user: on a phone screen and radio use are also big power draws.
Intel still has three problems: it is still playing catch up in the lower power game so while Silvermont has been announced, the S4 already has big.Little on the shelves; ARM manufacturers are catching up in process technology which is driven by volumes; Intel still wants to charge a lot more for chips than competition-constrained ARM vendors. It will obviously offer sweeteners such as marketing subsidies to any manufacturers to effectively lower the price per chip. But, as the continuing dearth of volume Atom-based phones would seem to indicate, it has yet to achieve volume.
I think that beefing up the architecture may be a gamble now that the phone market seems lost. This could easily hit revenues if computer and server manufacturers reckon they can do the same work with Atoms that they have been thinking about doing with ARMs because it will cannibalise sales of even beefier but much higher margin chips. There may also be reasonably large market in Windows Pro tablets, now that RT is effectively dead and the Start button is coming back, as drop in replacements for notebooks which are now standard business issue.
I'm not a pundit but I think we can expect the next earnings call to factor in lower margins, in a similar way that the lower margin sales of the I-Pad Mini have hit Apple's business.
If you read the pressrelease you comment about here, they are talking at least 50% of time about execution performance. But ARM is strong because it is dirt-cheap( $2 or less per smartphone/tablet CPU), consumes very little power (so battery of a tablet lasts whole day) and can be combined in almost arbitrary ways on the same chip. To top it up, you can play 17 die manufacturers against each other to push prices down.
So, free management consultation to Intel would be: attack ARM head on: lower power consumption, lower prices, Intel fabs etching onto the SoC whatever the customer desires. Allow customers to manufacture the intel design in other fabs (yeah, that's apostasy, but sometimes you need to convert religion to stay in business). If you chose to not do that, then my dear Intel, your days are absolutely numbered.
On the other hand, if sleazebag intel were Garbage-Collected, it would be a time to cheer. Maybe there's justice after all !
"So, free management consultation to Intel would be: attack ARM head on: lower power consumption, lower prices, Intel fabs etching onto the SoC whatever the customer desires. Allow customers to manufacture the intel design in other fabs (yeah, that's apostasy, but sometimes you need to convert religion to stay in business)."
Worse. It lets the customers get a real handle on the real cost of mfg of Intel processors.
The Intel line has been "Yeah, it's expensive, but it's complex and look what you get for the money (like running Windows)."
Look at how hard Intel fought to stop anyone second sourcing their chips after the 386 (which only happened because the Pentagon demanded a 2nd source for such a key component). Changing to a name, so they could copyright it and sue anyone else from calling their chip the same.
This is a company obsessed with controlling exactly what you are allowed to see and know about its product.
The fall of the netbook is largely down to Microsoft, not Intel. I have an EeePC 1000 running Windows 7, and it's usable. How? You upgrade the RAM to 2Gb and the disk to an SSD. I think it was Microsoft who insisted that manufacturers sold them with a maximum of 1Gb RAM and a hard disk (though of course, SSDs weren't cheap enough for the size needed until recently).
More cores and a bit more speed will be nice, but for anyone wanting to run Windows on something that we may as well still call a netbook (small, light in weight, long runtime on battery, and cheap) the keys are SSD and plenty of RAM. In the latter connection, I hope that Intel won't hamstring the new CPUs with too small a physical address space.
Microsoft didn't limit Netbooks to 1GB of RAM. They limited Win7 Starter licenses to systems shipping with only 1GB of RAM. OEMs weren't required to to ship Netbooks with Win7 Starter, they chose to use Starter rather than Home or Pro because that created a new price niche that they could fill.
There was absolutely nothing to stop HP or Dell or Lenovo or Toshiba or Samsung or ........ from building an Atom based system with an 11" screen, Win7 Pro, 4GB or RAM and a 256GB SSD, and selling it for $450, except for the reality that it would have been a complete and total flop. The "netbook" market segment existed because it could be produced at a price just below what you could build a basic laptop for - you could get a "netbook" for $250-$300, or a "laptop" for $300-$400, so a $400 "netbook" just didn't fit into that space, as far as the OEMs were concerned.
And whether you like it or not, the OEMs that count units shipped in Millions are the only ones who can turn a profit on a sub $300 netbook/notebook/laptop, so it doesn't make commercial sense to them, it doesn't get made.
So the question is that enough relative to ARM?
This is Intels SOP. When in doubt throw transistors at the problem.
It's what they do. But will it work? But look at Intel's weaknesses.
More expensive than ARM. Don't like integrating non Intel IP onto the chips (so your expensive Intel processor cost is multiplied by the # of additional (big) chips you have to stick on your board (and design for). And of course the Pentium instruction set is just huge
Here's the thing. How many people develop in assembler?
Because for most people these days what processor you support is all about the tool chain and do you need Windows "compatibility" (WTF that means given the games MS have played with their various "mobile" and CE OSes and apps over the years).
So right now "meh."
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
