Deep inside ARM's new Intel killer

ARM has swung a one-two punch at Intel's plans to muscle in on the smartphone and tablet space that's currently dominated by the plucky chip designers from Cambridge. At press soirées in London and San Francisco on Wednesday, ARM announced both a design for a tiny new chip, the Cortex-A7 MPCore, and a system-on-chip scheme …

[Anonymous Coward]

Intel's only way to success

is to buy Arm.

IT is really clear that the ARM engineers are prepared to think outside the box and come up with creative solutions. Whereas the intel designers always have to start with an x86 core. Because that is a (to be frankly honest) a crap design they are trying to compete with one arm tied behind their back.

Intel should either buy ARM of if they persist with their own cores then strip off the legacy x86 crap. Underneath, there is a pretty decent RISC core just waiting to be used. It would get the transistor count down by a goodly amount.

That is IMHO, probably a few steps too far for Intel thus leaving buying ARM the only option.

[Ken Hagan]

Re: a goodly amount

"Underneath, there is a pretty decent RISC core just waiting to be used. It would get the transistor count down by a goodly amount."

No recent Intel chip has devoted more than a few percent of its die area to instruction decode. That's not a goodly amount. That's an afterthought. ISA ceased to matter at the desktop/server end of the market over a decade ago. (Ironically, it probably started to fade at the same time as Intel were pushing a new ISA -- ia64.)

[bazza]

Poor old Intel

ARM have moved the goal posts once again. Intel are going to have to do something pretty amazing to make x86 anything like relevant in the mobile space, something they've completely failed to do over the past few years.

I don't think that there is any physically feasible way for Intel to make x86 compete with ARM in terms of performance/Watt. x86 is just too inefficient in its use of transistors. If there was a way you might imagine that Intel would have found it by now, but they haven't. And afterall, Intel have always been masters of silicon processing first, architectural geniuses* second. I can't see their current strategy paying off.

Perhaps one way in which Intel could have an orthogonal but effective strategy is to get in to screen technology in a big way. That's definitely an area where *big* power savings could be made, ARM might do a 20GFLOPS core that takes no power at all, but it's useless without a screen. Alright, so screens are not Intel's core business, not anywhere close, and there's a lot of competition in that field already. But they're wasting their cash right now so they might as well spend it on something that may bear fruit. It might also prove to be a company saver when Intel wake up one morning and discover that ARM have pinched the server market from under their noses too. With Microsoft porting Windows to ARM, and the LAMP stack already quite well established on ARM, I can't see Intel hanging on to servers for much longer.

[Ken Hagan]

Did you look at the performance graphs of the two ARM cores?

If you design a chip to have a high maximum performance, its lowest power state is still going to look rather greedy compared to a rival chip that cannot compete on performance. That's a basic rule and applies even to different breeds of ARM chip.

If Intel want to compete at the ultra-low power end of the market, they will be able to do the same things as ARM. Conversely, if ARM ever want to compete with Intel's offerings for server blades, they'll find that their performance per watt starts to fade.

The laws of physics are the same for both companies.

[Lance 3]

Atom

Then what was the Atom for? It sucks power down like no tomorrow and yet it doesn't outperform ARM either.

[Ken Hagan]

Re: Atom

A fair point. To be honest, I think the Atom was "for" filling a whole in someone's product matrix whilst the company decided whether they were serious about the low power market. It's still not clear to me that Intel *are* serious. Remember, they made ARMs for a while and then sold the business on.

Maybe they want to treat ARM like AMD -- someone to point the regulators at every now and then, whilst they (Intel) carry on coining it in the server market. Perhaps that will change if (or when) ARM deliver a chip that looks good in a blade. Then again, they've had a decade to do that (ever since Linux took root in server farms) and don't seem to have got around to it.

[Lance 3]

They sold the ARM division to focus on x86 and to bring it to where they were selling ARM. So they decided to enter the market and failed. They failed at the TV market as well. Intel wants x86 everywhere and it is just not going to happen. Even the large Xeon trounces the Itanic 2. Do you really think Intel wants to continue to sell it? They want everything to be x86.

[AndrueC]

It's useful for low power conventional computing. My server runs on an Atom. That's a bog standard Windows 7 installation running SqueezeServer, VPOP3 (email), FTPZiller and TVersity. It does a perfectly acceptable job yet consumes less than 10w. According to the manufacturer it should be no more than 5w when it's idle.

There may well be CPUs that consume less power - but not many of them run Windows and allow me and others to leverage decades of experience.

[I_am_Chris]

Go ARM

Great to see a UK tech company doing so well. <clap>

The gov must take note and grab the knowledge economy by the horns and invest in STEM industries. From school upwards...

[peter_dtm]

$diety's sake

don't even let the gov know there's a successful UK company - they'd find some one to stuff it up - and if they didn't manage it they'd surely tell Brussels. And they *know* electronics is a German thing........

either way they'd be stuffed

[Alex Walsh]

Why?

ARM's business model means they make money, but not a mad amount of money. And the skill is obviously in the staff, not the company. Intel have probably spent more trying to compete than they'd need to spend buying either ARM or it's staff many times over.

[bazza]

Because

Intel can't buy ARM. Just like Apple can't, nor Qualcomm, TI, Marvell, or anyone else. The competition regulators all over the world would have a monumental fit (or at least they ought to).

If any of the mobile players succeeded in taking control of ARM they would gain a de facto monopoly position, or at least the strong impression of one. An actual monopoly would clearly not be good for any end user whatsoever.

I would argue that the richness and diversity of the mobile market is traceble solely to the way in which ARM have licensed their CPU designs with an even and fair hand to all device manufacturers. They (perhaps accidently?) created a level playing field in which many manufacturers could thrive.

ARM survive because they themselves don't actually make anything, or sell anything directly to the public, so they're not really operating a monopoly either. And by being obviously modest about their licensing fees they cannot be accused of exploiting their dominant position either.

[Anonymous Coward]

Careful what you wish for !

Hate to bring politics into a technology debate, but the current UK Gov, is interested in CASH! Cash for its donors! Cash for its ex-members/members and Cash for their retirement! If the could get ARM sold for a 10% handling fee. They would shaft this brilliant UK company in an instant. We haven't forgotten INMOS and the transputer!

[Victor Ludorum]

That's all very well, but...

Can anyone explain why ARM's share price dropped by about 5% this morning? I can only assume someone somewhere thinks ARM won't get this working. They've had nearly 25 years experience, I think they'll make it!

*Off to buy some cheap ARM shares, long term investment and all that...*

V.

[Anonymous Coward]

Often positive news from ARM causes a decrease in the share price - I have seen this affect in other tech companies too.

[SoftFox]

Re: That's all very well, but... #

As with all company PR announcements they are obviously pre-planned so the market tends to adjust prices to them before they go live. This restricts the amount of price fluctuations that happen.

In ARMs case would seem it went slightly down which would indicate that the market anticipated it pretty well.

[Tomato42]

ARM Laptop

I only ask one thing: where's an ARM-based, Linux-running clamshell laptop, not a netbook, but a ultraportable or 14"?

[bazza]

Tiny market

"...Linux-running..."

Without passing judegement on Linux per se, my guess is that the perceived market demand for such a thing is simply way too small to interest any large manufacturer.

But market perceptions have a nasty habbit of being wrong. Remember IBM's estimate that there would only ever be a need for 5 or 6 computers in the whole of the US? What a mistaka to maka!

I'm hoping that whatever MS are upto with ARM will result in an open ARM platform just like the x86 platform is at the moment. MS bought an ARM foundry license, (a *lot* of wonga) so it seems they're hell bent on building an ARM platform of some sort.

Microsoft, and latterly Linux, benefitted enourmously from IBM's architectural openness that spawned the whole PC ecosystem. MS have some form in this area too. The PC'97 -> PC2001 series were sort of along the lines of an architectural spec that served to standardise PC hardware. That definitely served MS's commercial purpose - they could sell more Windows licenses as a result. It also helped other things like Linux too for the same reasons. Maybe, just maybe, MS have decided to try pull the same trick with an architecture based on ARM.

MS commercial interests to do so is that they could spawn a whole new major round of platform evolution just like IBM did back in the 1980s with the first PC. I think that their purchase of an ARM foundry license is evidence that they're aiming to create a whole new ecosystem, from server -> desktop -> mobile.

Whomsoever successfully pushes ARM into the server market stands to make a shed load of money. Those datacentre operators are desparate to reduce their electricity bills, and they'll spend big on hardware to do so. Energy costs, as well all know very well indeed, are king.

I wouldn't mind betting that MS have worked out that by defining the hardware they'll be in a good position to sell a very large number of software licenses too. The implication is a substantial replacement of the PC computing world as we know it, not just annual incremental sales. All that MS and the hardware vendors have ever sold they get to sell again, in ARM form.

Of course, if Intel actually stumped up a decent low power chip that would mean the world would just continue with incremental license purchases rather than complete replacement. That wouldn't make MS anything like as much money. Essentially I'm arguing that Intel's failure to produce a proper low power chip is an enourmous once in a lifetime commercial opportunity for MS.

It's a very large market to aim for, it must surely be tempting for them to go for it. And if it makes their mobile strategy work too, so much the better. And where there's servers, there'll be desktops and laptops, and also Linux devs who'll inevitably work out how to penguinise it.

Anyway, assuming that I'm foretelling with accuracy, your quest might be successfully pursued merely by sitting tight and waiting for it to happen, perhaps sooner rather than later. The only loser would be Intel, and they won't be happy at all.

[Voland's right hand]

This will make some very hairy scheduling algorithms

I do not want to be the guy who is blamed to write the scheduling algorithm for this.

While nice on silicon, the OS support for this will be hugely entertaining.

[SoftFox]

Re : This will make some very hairy scheduling algorithms

The CPU changeover will be transparent to the OS so no extra coding required. The cores themselves have been designed to be identical from an external point of view so its just like any other quad/dual processor

[Ken Hagan]

Re: OS support

The impression I got from the article was that it would be little more than a context switch. You need to save the on-CPU state and then send a message to "something" to shut down the current processor and fire up the other, whereupon the saved context is re-loaded and continues execution. I'd have thought the OS support would be trivial.

[Ken Hagan]

Ahh, takes me back...

ARM seem to be doing what AMD did in the late 90s: giving Intel a serious kick up the backside. Last time round, AMD killed ia64 and probably hastened the demise of the Pentium 4. Intel were saved only because they had apparently unlimited fab capacity and several skunkworks teams hidden in the basement. They still have the fab capacity, but do they still have the heretics?

[fearnothing]

A while back I was doing some modules on an MCSA course and the tutor said he'd been speaking to one of Intel's engineers back when AMD was wiping the floor with them using the new Athlon line. Apparently this engineer wouldn't go into specifics, but just said with a sort of smug confidence, "Wait two years."

That was in 04. Less than two years later, the Core series came out and Conroe was selling like hot cakes. That's one hell of a skunkworks team!

[bazza]

@fearnothing

The Core series was nothing more than a couple of Pentium 3's stuck on to the same package with a miserable excuse for a front side bus to join them together. Hardly any work was involved at all, and it was seen at the time as something of a desparate measure.

The fact that it took them 'two years' to accomplish that is not very impressive at all, frankly. In that timescale *the* Skunk Works team were quite capable of delivering entirely new types of aircraft such as the F117a prototype, the U2 (the A12/SR71 took a little longer, but not much).

However, Core was a marketing success which as is so often demonstrated far more important than technological success. Intel were able to exploit the fact that most customers would open the spec, note "dual core" and look no further. However, with power consumption becomming ever more important the customers are likely to take specs more seriously.

All it will take is some large-ish datacenter operator to fit out with ARM based hardware and give it a go. Not that straightforward, granted, but LAMP is LAMP be it on x86, ARM, etc, and MS are busily putting all sorts on to ARM so it's getting easier. It is quite likely that that datacenter operator would achieve a major power saving. And that is a major cost saving. And that is a major profit increase. And that *will* get noticed by others, because their shareholders will start complaining. And that will get Intel into big trouble, because they can't respond without tossing x86 in the trash bin and starting again.

[Dr Gerard Bulger]

Two Processors

Nothing really is new. Anyone remember the Wireless Worl kit PC that was about in late 80s.. Sinclair ZX time? I had great fun with one. That machine had Z80 chip and a reverse polish co-processor. The concept whent nowhere of course as pretty graphics is what the market needs.... and not sure what happend to reverse polish either.

[Chika]

If you want to go there, consider that ARM themselves came out of another British company called Acorn who weren't averse to doing similar things with processors. Consider the heart of the Acorn Electron, which was a mashup of various processor bits found separately in the old Beeb, or a similar setup of one of the earlier ARMs, the ARM250, found in certain Acorn Archimedes models around the time that Acorn were in the process of spinning ARM Holdings off.

Nope. Nothing new under the sun^H^H^Horacle.

[Mage]

Maybe if Intel put 28nm versions of the old 386SX and the Pentium pro on one chip they might do it.

A billion transistors? Intel has got too bloated concentrating on simply server/high end desk top performance over all else.

They had to step back in time to do the Atom (for low power), ironically the 1st Atom based netbooks about 1/2 the speed of a 10 year old laptop.

[Anonymous Coward]

Dreadfully-phrased, but hilarious

"For years, Intel has been trying to push its products down into the exploding smartphone and mobile-device market."

Oh dear me.

[Beachrider]

You can root for the home team...

Nothing wrong with Britons rooting for British industry. Analysis from 'homers' isn't always the best analysis, though. I like ARM, for its ability to build low-energy processors. It could be that this development will evolve into a competitive jewel.

[Anonymous Coward]

Re: You can root for the home team...

Rik is Californian, so I am unsure if this qualifies as home team.

[rav]

Why?

Why is it that ARM, by building miniscule, painfully slow and low perfroming chips, is considered an Intel killer and AMD who builds cutting edge technology and keeps Intel honest and can't get any respect?

[bazza]

Because

They're not especially slow or low performing. They tend to be well matched to the things that people actually want to do. That's why they've been so successful in smartphones in particular.

What interests me and many others is whether that right-sized characterstic translates over to the datacenter. Your average Intel chip is doing a hell of a lot of things. But to make economic sense it has to, it's burning through 100+ Watts. So could you do the same amount of work with a few ARM chips? If the answer is yes then you use them, because you'd likely save a lot of power. Even if it took 10 ARMs that'd still be OK - that'd be about 20 to 30 Watts, saving 70 Watts. 10 ARMs would be 10 cores at 1+ GHz - not bad really when you think about it.

[Ken Hagan]

Re: Why?

Because people like stories.

You can't run Windows software on ARM. That makes ARM "revolutionary" compared to AMD's "normal", and that's enough of a hook for the great unwashed to get excited about the story. This isn't going to change anytime soon, since we've already been told that Win8 will not include x86 emulation for your existing apps. That makes it "just a phone thing" for the 99% of potential customers who have spent more on software than on hardware.

Expect to read a lot more stories in the next couple of years about how revolutionary ARM are just about to overthrow Intel's dominance of the hardware business, in the same way that Linux has been just about to overthrow Microsoft's dominance of the OS business for the last decade or so. As I said, people like stories and they cling to them in the face of all experience.

I expect *Intel* give AMD more respect than they give to ARM. The former compete head to head and affect profit margins in their existing market. The latter are merely blocking Intel's entry to a new market. Annoying, but not actually painful.

[Charles 9]

Thing is...

The mobile and tablet markets threaten to be a market DISRUPTOR. The only thing that would scare a market incumbent more than a market competitor is a market disruptor. Because that means the ENTIRE market, not just your share of it, is under threat. Put it this way. What if mobile and tablet computing were to take over the home. Instead of playing games and whatnot on PCs on desks, they took control of their tellys and played games off their tablets through OnLive or the like? Intel can't get a chip in edgewise in the phone and tablet markets, and if people buy more phones and tablets and fewer home PCs, their chip sales start dropping.

And now the enterprise market is looking into tablets. iPads are starting to appear in offices and so on, since they more closely resemble Star-Trek-esque PADDs and make the "paperless office" concept of the 80's more realistic (since you're not just reading the document, you're HOLDING it). And with cloud computing taking off, there is another chance for the thin client (which doesn't require a specific chipset) to take over the office cubible. There goes more Intel chip sales.

About the only stalwart left for Intel is the datacentre, where power efficiency is a key factor but throughput still rules (time is LITERALLY money, aka profits, here--saving energy money doesn't help if you it's offset by lost throughput profits). This is where ARM or some other company of its like still hasn't established itself. If someone can pull off Intel-level amounts of performance per second and STILL draw less power, then Intel better start praying.

And as the article notes, don't expect Microsoft to help Intel this time. With their announcement of supporting ARM in Windows 8, Microsoft is clearly hedging its bet this time.

[xperroni]

Those big.LITTLE supporters...

Aren't a number of them also "supporting" WAC?

Just asking.

[Anonymous Coward]

Hasn't HP just bought an ARM server outfit?

Interesting times...