Intel Shifts Might To Mobile

CWmike writes "After years of dominance in computer chips, Intel now is chasing the mobile chip market and trying to redefine its future. During Intel's financial analyst meeting Monday, CEO Paul Otellini announced that he is refocusing the company, moving its 'center' from PC processors to processors for the burgeoning mobile market. 'I think Intel recognizes that they absolutely have to get a win here,' said analyst Rob Enderle. 'All the activity is in mobile. A post-PC era would be a post-Intel era if they don't get a beachhead established.' Earlier this month, Intel made a move in this new direction when it unveiled its new 3D transistor technology that is expected to position the chip maker to grab a piece of the mushrooming tablet market."

Comments

little late

[Osgeld]

to the game, everyone and their grandmother is fabbing arm chips under their own flag for their own use, though I wish you luck intel

Re:little late

[fuzzyfuzzyfungus]

Quite possibly too late to x86-ize the market, and capture the sort of margins that they have historically enjoyed; but being the man with the best fabs in town doesn't sound all bad when the rest of the town is guys cutting each other's throats over generic ARM SoCs fabbed on assorted unexciting processes...

Re:little late

[UnknowingFool]

Traditionally Intel didn't really pursue this market because of the low margins. Also ARM was far better at power consumption/cost whereas x86 was better at performance/cost. They can't ignore the market anymore. Tablets are predicted to sell 40-50 million units this year. And every one of them whether iPad, Xoom, PlayBook, whatever will not use an ATOM chip. While tablets can't replace laptops outright, they can replace enough functionality where a person buys a tablet instead of a 2nd computer.

Re:little late

[TheRaven64]

Not just tablets. I have an ARM-based laptop. It's painfully slow on big compile jobs (LLVM takes over 5 house to compile - ouch!), but for FireFox and OpenOffice it's fine. It's cheap, light, and has a decent battery life. I absolute terms, it's much slower than x86 machine, but for a lot of users it would be fast enough.

Re:little late

[blair1q]

You forget, perhaps, who Intel is, and what they have.

Everyone and their grandmother may be ordering ARM-based chips from the contract fabs, but Intel does its own design and fabrication.

Their economies of scale and vertical efficiencies are not something that the ARM world can stay ahead of for very long.

If Intel has decided that there's enough profit in this sector to make it their major focus, their monthly spending on it could outstrip everyone else's annual expenditure.

BTW, they've been in mobile before. It just wasn't big enough for them to make real money at it. Now is a whole new situation. Mobiles are a lot more like computers than they are like phones, so putting more computer-like CPU cores into them is a logical idea.

ARM needs to start playing catch-up just to stay in the race, even though it's ahead in the early laps.

Re:little late

[blair1q]

They've been putting a little attention on mobile, and playing a little poker with the market. Now, clearly, they see that desktop is about to become a lower tier, and mobile will be the major sector, and server will be the hidden half of the mobile sector.

Their ability to work at low power and small form-factor is improving to competitive levels, and they can work past any deficiency they have in those areas.

So now they're actually putting their major focus on mobile. The 800-lb gorilla just entered the room.

Re:little late

[serviscope_minor]

They can't ignore the market anymore.

Why not?

And every one of them whether iPad, Xoom, PlayBook, whatever will not use an ATOM chip.

But they all use online services. Those online services need big, fast computers to run them. I'm sure Intel has benefitted massively from the massive use of non intel devices to access intel servers.

Re:little late

[UnknowingFool]

Why not?

Money and marketshare? Like I said in the rest of my statement, consumers are starting to buy tablets instead of a 2nd computer. Right now every tablet sold is not using Intel chips. Ignoring the market means they are ignoring potentially billions of dollars in the future. And that's just tablets. Netbooks, smartphones, and other devices are being powered by ARM. As these devices become more plentiful, that's another source of potential revenue and market that won't include Intel.

But they all use online services. Those online services need big, fast computers to run them. I'm sure Intel has benefitted massively from the massive use of non intel devices to access intel servers.

I never said that Intel can't still sell computers. I said that Intel can't ignore the rapidly growing markets of mobile devices.

Re:little late

[yuriyg]

While tablets can't replace laptops outright, they can replace enough functionality where a person buys a tablet instead of a 2nd computer.

Absolutely not true. Most users will be more than OK with a tablet. These things are designed for web, e-mail, and casual games; and let's face it, this is all the average user wants to do. Spreadsheets and word processing? Save that for a desktop at work. Getting back to topic, this is not to say that Intel won't be able to capture the market. With the best r&d in the market, I believe they will soon become number one chip maker in the mobile world.

Re:little late

[Glock27]

Fairly good analysis, especially in that both absolute prices and margins have dropped considerably.

What would have been a $300,000 server not all that long ago is now a $3,000 server... ;-)

Re:little late

[marcosdumay]

Either move is risky. If Intel changes its focus to low power it may lose its strengt at the hight speed market, and may, or may not open another market for itself. If Intel doesn't change its focus it may have their market reduced under their feet, and in a couple of generations power consuption may become more important than speed even at the datacenters.

But then, you are quite too far out there...

"Their economies of scale and vertical efficiencies are not something that the ARM world can stay ahead of for very long."

They have almost no market right now, how can the entire market not compete with them on economies of scale? Also, vertical efficiencies means that their customers won't be able to fine tune the processors to their needs, that is ok if you have a little power to spend on circuitry you won't use, but we are talking about mobile.

Re:little late

[UnknowingFool]

The main reason I said 2nd computer is that you still need a computer for an iPad and other tablets. So instead of a new laptop in addition to a desktop, consumers might opt for a tablet. Instead of a 2nd laptop, they may opt for an iPad.

After all...

[TheRedDuke]

Who needs Xeons and i7's to run servers when everyone has a super-powerful smartphone now?

Oh, wait...

Re:After all...

[h4rr4r]

Considering the power savings a couple hundred ARM cpus might make for a decent server.

Re:After all...

[the linux geek]

A couple hundred ARM CPU's still can't deliver decent single-thread performance, which is more important for servers than a lot of people seem to think. If that wasn't true, the Oracle Niagara chips (128 threads @ roughly 200MHz) would be doing great right now instead of bleeding market share.

Re:After all...

[TheTyrannyOfForcedRe]

1. SeaMicro makes ultra high density servers out of Atom chips. Here's one with 512 Atom cpus: http://www.seamicro.com/node/102

2. SeaMicro's offerings are well received so Atom must be fast enough for server use.

3. The faster ARM chips are neck in neck with Atom on integer performance. Their floating point performance sucks but we're talking about regular servers not HPC.

Conclusion: ARMS are fast enough for server workloads.

Re:After all...

[serviscope_minor]

1. SeaMicro makes ultra high density servers out of Atom chips. Here's one with 512 Atom cpus: http://www.seamicro.com/node/102

I see that they pitch it as replacing 40U's of dual quad cores (8 core machines).
You can now buy 48 core machines easily. That's 7U's worth of servers. And the seamicro one is 10Us. So, it's actually less dense than other offerings (thats's according to SeaMicro).

In terms of power, a 48 core machine draws a bit over 1kw on full, so 7kW for the rack. The seamicro machine has 3 (+1) power supplies. Given the 240V input, that could be anything up to 9kW, ut they don't give full load figures.

So, it doesn't actually seem all that compelling, to be honest.

Re:After all...

[h4rr4r]

If they sold them at reasonable prices they might do pretty well.

XScale

[afidel]

So why did they sell XScale off a few years ago? I mean this is the same CEO that pulled the last refocus off and obviously missed so I wonder why the board and investors think he'll be able to pull this one off without seriously missing the mark again?

Re:XScale

[lotzmana]

I think it was sold because of pride. How could a CPU giant license designs from puny ARM from England?!

-- http://bashrc.sourceforge.net

Re:XScale

[blair1q]

What was the smartphone market like a few years ago?

What is it like now?

Intel made cash selling Xscale, made more cash not trying to leadership the sector, and is using that cash to get back into it and take it over. And anyone doubting they can dominate it just isn't paying attention to cost and scale.

I don't know if they were smarter or luckier, but they certainly were both to some degree.

Re:XScale

[tlhIngan]

I think it was sold because of pride. How could a CPU giant license designs from puny ARM from England?!

Actually, they had a microarchitecture license that let them implement the instruction set any way they wish. Most licensees are just given the RTL and that's it. ARM's been a bit more generous lately since then - letting Qualcomm (Snapdragon) and Apple to also purchase such licenses.

Intel came about it through Compaq lawsuit settlement who got it via purchasing DEC. They inherited the StrongARM, then their own chip designers created XScale. They then sold the entire division to Marvell.

To this day, though, StrongARM/XScale remains the truly independent reimplementation of ARM and ARM's tools all mention it separately from ARM's designs.

Rob Enderle

[Citizen of Earth]

Rob Enderle of "SCO's gonna win!" fame.

Re:Rob Enderle

[aristotle-dude]

Rob Enderle of "SCO's gonna win!" fame.

Also, Apple's "Insert iProduct Here" will fail.

He also predicted that HD DVD would beat out Blu-ray just before Warner Brothers announced that they were going Blu-ray exclusive. Rob was on the payroll of Toshiba as a "consultant".

Re:Rob Enderle

[Jason Earl]

I think that it goes without saying that Enderle calls more wrong than he calls right. However, in this case it is hard to argue with him. Microsoft is apparently going to release a version of Windows 8 with ARM support, and between tablets and smartphones people are doing more and more of their computing on devices without Intel inside.

Intel needs a low power processor that competes with ARM badly, and Atom is not getting the job done.

not a matter of instruction set

[lotzmana]

People think Intel's purpose is to impose the x86 instruction set and also that the only culprit that keeps them from making a successful product is the overhead of that very x86 instruction set. I don't believe it.

The interpretation and translation of instructions is some constant number of transistors, the rest of the architecture is moving ahead. There will be a moment when brute force alone, the supremacy of the fabs, will win the race.

Another factor is that when you license ARM you can customize it. You can't license and customize Atom CPUs. Intel is at the moment kept back by a combination of factors

-- http://bashrc.sourceforge.net

Re:not a matter of instruction set

[TheRaven64]

The interpretation and translation of instructions is some constant number of transistors, the rest of the architecture is moving ahead

Not really. One of the things the VirtualPC team at connectix discovered was that a large number of x86 instructions have side effects (e.g. setting condition flags) that, 90%+ of code ignores. In a hardware x86 implementation, you have to burn energy computing these.

There are lots of other things that make x86 harder to implement efficiently, for example the lack of predicated instructions. ARM can do short conditional statements without needing branches, which means that it can get away with a simpler (and therefore less power-hungry) branch predictor.

It's not just a case that you translate x86 or ARM into more or less the same RISCy set of micro-ops and then run them on similar hardware - the ISA forces certain design decisions all the way along the pipeline.

You can't license and customize Atom CPUs

Yes you can. You've been able to for over a year. Intel will even fab the customised SoC for you. As yet, I don't know of any company that has chosen to do so, however.

Re:not a matter of instruction set

[thammoud]

On Android, it is not even relevant given that the vast majority of apps (Java) do not use native code.

Re:not a matter of instruction set

[gnalre]

The parent has a good point. The concentration has been on power consumption as the reason that Intel are not in the mobile game. Buts thats only part of the story.

The ability to take an arm core, customise it to your specific requirements and therefore differentiate yourself from the rest of the market is just as important. Intel are to expensive for low end products, but the high end market don't want to be clones.

Complete headlines please?

[guspasho]

Is it really that hard to write a headline that includes enough words to make it clear? This isn't print where more words mean more space and ink or a smaller font size for the same space.

Re:Rob Enderle - Why quote him????

[xzvf]

Maybe we should have an ask slashdot where authors explain why they think, what he says is valuable or interesting and worth quoting.

If you can't beat them join them

[edxwelch]

I don't understand why they're even bothering with Atom. It's about 100 times more power hungry than an ARM. But ARM will license their chips to anyone, so why not just make an ARM? With their new 3d process technology they would have the lowest power consuming ARM chip on the planet

Re:If you can't beat them join them

[RightSaidFred99]

It is not 100 times more power hungry than an ARM. In fact in the phone and tablet spheres it's competitive on power. Not class leading, but let's say 2X more power hungry than ARM (to be generous to ARM). It also performs better. This gap is only going to close (and get turned on its head) as Intel puts more focus on mobile in terms of design resources and top-end FAB process access.

It's laughable hubris to claim Intel doesn't have a chance in mobile with Atom, in fact I'd say it's wrong to say they have anything less than a very very good chance.

Re:If you can't beat them join them

[edxwelch]

I don't know about that. I've seen benchmarks that show Atom using 50x more power in idle mode.

ARM vs x86 review

[IYagami]

http://vanshardware.com/2010/08/mirror-the-coming-war-arm-versus-x86/

Conclusion

The ARM Cortex-A8 achieves surprisingly competitive performance across many integer-based benchmarks while consuming power at levels far below the most energy miserly x86 CPU, the Intel Atom. In fact, the ARM Cortex-A8 matched or even beat the Intel Atom N450 across a significant number of our integer-based tests, especially when compensating for the Atom’s 25 percent clock speed advantage.

However, the ARM Cortex-A8 sample that we tested in the form of the Freescale i.MX515 lived in an ecosystem that was not competitive with the x86 rivals in this comparison. The video subsystem is very limited. Memory support is a very slow 32-bit, DDR2-200MHz.

Languishing across all of the JavaScript benchmarks, the ARM Cortex-A8 was only one-third to one-half as fast as the x86 competition. However, this might partially be a result of the very slow memory subsystem that burdened the ARM core.

More troubling is the unacceptably poor double-precision floating-point throughput of the ARM Cortex-A8. While floating-point performance isn’t important to all tasks and is certainly not as important as integer performance, it cannot be ignored if ARM wants its products to successfully migrate upwards into traditional x86-dominated market spaces.

However, new ARM-based products like the NVIDIA Tegra 2 address many of the performance deficiencies of the Freescale i.MX515. Incorporating two ARM Cortex-A9 cores (more specifically, two ARM Cortex-A9 MPCore processors), a vastly more powerful GPU and support for DDR2-667 (although still constrained to 32-bit access), the Tegra 2 will doubtlessly prove to be highly performance competitive with the Intel Atom, at least on integer-based tests. Regarding the Cortex-A8’s biggest weakness, ARM representatives told us its successor, the Cortex-A9, “has substantially improved floating-point performance.” NVIDIA’s CUDA will eventually also help boost floating-point processing speed on certain chores.

Re:ARM vs x86 review

[RightSaidFred99]

ARM Cortex-A8 matched or even beat the Intel Atom N450 across a significant number of our integer-based tests, especially when compensating for the Atom’s 25 percent clock speed advantage.

Disqualifies anything else written, as this is idiotic. You don't compensate for clock speed advantage, it's part of the definition of a CPU's performance. If a 5GHz CPU performs 1% better than a 3GHz CPU, the 5GHz CPU is faster. Period. Of course, one would also take into consideration power usage and other metrics, but those are not performance. One might say "this 5GHz CPU is faster than the 3GHz competition, but uses twice the power". But one doesn't "compensate" for a clock speed advantage. An advantage is, well...an advantage.

Re:ARM vs x86 review

[Zan Lynx]

Don't be stupid.

If the article is compensating for clock speed then they obviously aren't measuring raw performance, but performance per clock. This is a useful metric in all sorts of ways.

Re:ARM vs x86 review

[0123456]

Look at a ATOM board. The large thing with the cooling... that's the northbridge!

You do realise that Intel only ship absurdly crappy northbridge chips with the desktop Atoms, right? My Atom netbook uses about a third as much power in total as the northbridge in my Atom home server does (something like 7-8W vs 20-24W, if I remember correctly).

ARM still has the lead on power, but it's nowhere near as bad as you make out.

Re:ARM vs x86 review

[WorBlux]

Rip the guts out of an old think book, and replace with those super-small ARM boards. A cluster in your briefcase. Of course a cluster doesn't work that well for a desktop system, but still the awesomess of it would shine.

Re:ARM vs x86 review

[lucian1900]

The review is stupid.

1. 1. NEON is very fast. Miscompiling your code for softfp doesn't mean ARM is slow.
2. 2. i.MX515 doesn't have to have slow memory, just this particular chip
3. 3. The Cortex A8 is a SoC, while this Atom isn't. Factor in the power used by the Atom's board too.

Re:ARM vs x86 review

[RightSaidFred99]

It isn't. It's useful in 0 ways, especially between entirely different processor architectures. Literally 0 as in zero ways.

Re:ARM vs x86 review

[TheRaven64]

NEON is very fast. Miscompiling your code for softfp doesn't mean ARM is slow

NEON is fine for single-precision, but (unless there's been a recent upgrade that I missed), doesn't support double precision. You have to fall back to VFP for double precision.

And, yes, performance does suck with Linux because Debian insists on defaulting to softfp mode. The worst thing about this is that it uses integer registers to pass floating point arguments, and you have a 15 cycle penalty for moving any value between integer and floating point registers.

Agree, many are underestimating Intel

[guidryp]

I agree, mobile has been playing second or more like third fiddle for some time. The mobile chips get the fab generation once the desktop/laptop chips are just about done with it.

If Intel was serious about Mobile they would be fabbing closer to the same generation as their desktop chips. But now that is exactly what the new roadmaps are showing.

I don't really think Intel will have much problem catching up on power usage and dominating on pure performance.

But the one stumbling block will be installed base in mobile where ARM is now the standard. I don't really see Apple moving it's mobile products back to x86. The OS isn't an issue, but all the native applications are.

Android is a mix of Java and Native code, so it would be a little messy there as well.

Re:Agree, many are underestimating Intel

[blair1q]

Apple is completely agnostic about architecture. Their move from PPC to x86 for Macs proved that for all time. Their constant flip-flopping on graphics is just more data. If they see performance and feature and price lining up in their favor with any chip, they will buy that chip. Jobs is everybody's favorite customer and nobody's friend.

Re:Agree, many are underestimating Intel

[Compaqt]

Well, wait: they can't just throw their desktop chips on mobile. They consume too much power (right?).

Re:Agree, many are underestimating Intel

[UnknowingFool]

The main reason they went with ARM instead of Atom is probably for the power efficiency rather than performance. That and they can customize their own chips with ARM whereas they can at best only collaborate with Intel on Atom. While Apple doesn't modify the core, they can modify everything around the core.

Re:Agree, many are underestimating Intel

[blair1q]

I'm old enough. So is Jobs. That crap almost buried Apple. And IBM wasn't nearly the only source of PPCs. Freescale would love to have the Mac account back.

Intel has a plan.

[symbolset]

ARM on the other hand has mobile products in the joyful hands of millions of happy customers. They have the support of Apple, Android device makers, HP WebOS and soon Microsoft. Every consmer electronics store in the world has mobile ARM products on the shelf and they are moving briskly - in a lot of cases rescuing vendors from a downturn in the economy where people aren't buying a lot else.

I am glad Intel has decided to tell us they intend to compete at this level. It indicates that at least they hear us now. But they have been promising that for five years and we have Atom that is not working in mobile. The desktops still burn watts like they are free. Laptops with Intel are thick, bulky, heavy and still take three minutes to boot. There is no credible tablet.

I want to believe, but after all these years and broken promises it is time to say "show me. If you have it, ship it. If you don't... then shut up and get back to work."

Re:Intel has a plan.

[Glock27]

Laptops with Intel are thick, bulky, heavy and still take three minutes to boot.

Except the Apple variety, of course... :-)

Re:Intel has a plan.

[blair1q]

Look at your iPad. Can you even determine without ripping it open what is inside? Most people have no clue nor do they care. It could be an ARM or a Tegra or a Llano or an Atom. (This btw is why Intel started the "Intel Inside" sticker campaign; so you know what it is that's really running your computer and they get credit for it.)

Apple can switch whatever they're using now for an Intel chip, and if 99.9% of the public even finds out it will be because of marketing, not because of the look or feel of the product.

So it means nothing at all that ARM is in the hands of millions. Very few people are brand-loyal in microprocessors; and those who are have an ulterior motive, not a performance or value motive.

Oddly enough...

[symbolset]

This is pretty close to nVidia's plan.

Re:ARM sucks

[TheRaven64]

Pigs? Well, maybe. My Core 2 Duo at 2.16GHz can compile LLVM in about an hour. My 800MHz Cortex A8 can compile it in about 5 hours. This makes the C2D about 5 times faster, but I'm doing a parallel build on the C2D, so in terms of performance per clock (for this specific workload), they're not far off - the C2D has two cores running at a total of 5.4 times faster. In terms of performance per Watt, the difference is more pronounced. the entire Cortex A8 SoC uses about 1W under load. The C2D uses well over 5.4W - even more including the chipset, so it's quite a bit faster per Watt. For floating point intensive operations, the C2D is a lot faster, but the Cortex A8 SoC has a DSP and a few other coprocessors that offload most common FPU-intensive algorithms - and still fit within that 1W power envelope.

Clock for clock, the A8 is slower than the A9, and dual-core 1GHz A9s are shipping now. The architecture scales up to 2GHz and quad core - that's more than enough for a large number of use cases.