Sunday, November 27th 2011
Wintel Alliance Slowly Crumbling, ARM To Eventually Rule The Desktop?
The writing has been on the wall for a while now, that the close relationship between Microsoft and Intel (and by extension AMD) is crumbling into dust. In fact, they have never really been the best of friends. It has been clear since Microsoft unveiled that Windows 8 would run natively on ARM processors that things would never be quite the same again. Apart from some niche server variants of Windows, which could run on Itanium and other processors, all the previous desktop versions, including Windows 7, have run on x86 (and x64 for the last 6 years or so) processors.
However, Microsoft is keen to increase its presence in the lucrative smartphone and tablet market, where it's not had much success so far, getting comprehensively trounced by Android and Apple. Microsoft would be happy to use an x86/x64 processor for this application, but here, the limiting factor is the energy source, the battery, forcing the entire device to consume very little power if it's to run for more than 5 minutes. To meet this requirement, processors based on the ARM architecture have met this need admirably for years, with excellent performance while the Intel x86 variants have not (see video below). This has lead Microsoft to forge a relationship with a new processor manufacturer, Qualcomm, who make their own variant of the ARM processor, called Snapdragon. In fact, the relationship is so close now, that Windows Phone 7 only runs on Qualcomm ARM chips.
Having Windows run on two processor architectures concurrently inherently puts them into competition, creating an uneasy, unstable coexistence (witness the death of the other architectures in the Windows server space) so it seems reasonable to expect that Qualcomm will end up competing head to head with Intel at some point. This should make for a very interesting situation, given Intel's strength and Microsoft's strength, which could be used to invest in Qualcomm to help it compete with Intel in the performance desktop market, which would be expensive and difficult in terms of R&D. Perhaps an alliance with AMD or IBM, given their design expertise could also be on the cards? Of course, the major show stopper for a full-on ARM onslaught into the desktop space is that "legacy" x86/x64 apps - which the whole world runs right now - either won't run at all, or will run poorly under some sort of emulator. The fact that all current ARM chips are physically optimised for low power rather than all-out data processing performance really doesn't help the situation, either.
For the moment, let's assume that this problem is successfully overcome, perhaps by porting various key apps over to ARM say. Due to the significant efficiency and performance improvements of the ARM architecture (see video below) x86 begins to be phased out, eventually disappearing. Now, where does this leave Intel? To go bust, obviously, as it can't sell any more x86 chips. No, of course not. Intel has had an ARM licence for years, so it seems logical that it would put its many superb data processing enhancement technologies into ARM chips, to create monsters that are capable of the blistering speeds we see today from x86 chips and then some. Intel really, really won't like this situation though. Why? Because at the moment, it's only proper competitor in the x86/x64 space is AMD, which conveniently for Intel, is some considerable way behind with its flagship Bulldozer architecture. One competitor. Easy to take care of. Could probably kill its x86 business if it wanted to, just by accelerating the performance of its chips by 50%. But then that pesky Competition Commission would start investigating…
However, the ARM CPU is made by literally hundreds of different companies, since ARM Holdings is a fabless company and makes its money by licensing the rights to make the processor. It doesn't take much of a stretch to see that some big hitter like IBM, who has similar expertise in building high performance processors (think PowerPC and POWER) could start competing with a high performance desktop variant of the ARM architecture. AMD will likely do the same, if they want to remain as a CPU manufacturer (they'd still have the profitable graphics card business to fall back on, so wouldn't die). Suddenly, Intel has lots of stiff competition from all sides and that extremely profitable niche that it has sat in for the last 30+ years due to licence exclusivity evaporates, perhaps eventually becoming a me-too commodity player with razor thin margins. Very painful, very humiliating, totally unthinkable. Maybe this is the real reason why Intel only ever made a half-hearted attempt with its Xscale ARM processors and the product line never really took off? It would have to literally be forced like this to make anything more of it.
So, you can see how it's completely in Microsoft's interest to move to ARM and absolutely not for Intel to do so. They now both want diametrically different things out of their long term relationship, so no wonder it's cooling off. Any bets on when the divorce papers will hit?
And now for that video. The short video below, originally found in an interesting geek.com article, compares a 1.6 GHz dual core Atom CPU in a netbook against a development board using a Cortex-A9 ARM CPU, configured as a dual core system, running at a mere 500 MHz. Yes, just 500 MHz. The results? Even with the netbook having a graphics accelerator and the ARM dev system not having one, the ARM was only slightly slower than the Atom! Of course, it consumed a lot less power than the Atom CPU too, which is critical. Note that this video dates from Jan 2010 and there's newer versions of both products now. However, it's still valid today, as the performance balance hasn't changed much between the two processor architectures. This is because the differences are inherent to them (x86 is hot and inefficient, basically) so it doesn't really matter how much each one is tweaked, the performance ratios will stay roughly the same.
Sources:
Sign On Sandiego, TechEye
However, Microsoft is keen to increase its presence in the lucrative smartphone and tablet market, where it's not had much success so far, getting comprehensively trounced by Android and Apple. Microsoft would be happy to use an x86/x64 processor for this application, but here, the limiting factor is the energy source, the battery, forcing the entire device to consume very little power if it's to run for more than 5 minutes. To meet this requirement, processors based on the ARM architecture have met this need admirably for years, with excellent performance while the Intel x86 variants have not (see video below). This has lead Microsoft to forge a relationship with a new processor manufacturer, Qualcomm, who make their own variant of the ARM processor, called Snapdragon. In fact, the relationship is so close now, that Windows Phone 7 only runs on Qualcomm ARM chips.
Having Windows run on two processor architectures concurrently inherently puts them into competition, creating an uneasy, unstable coexistence (witness the death of the other architectures in the Windows server space) so it seems reasonable to expect that Qualcomm will end up competing head to head with Intel at some point. This should make for a very interesting situation, given Intel's strength and Microsoft's strength, which could be used to invest in Qualcomm to help it compete with Intel in the performance desktop market, which would be expensive and difficult in terms of R&D. Perhaps an alliance with AMD or IBM, given their design expertise could also be on the cards? Of course, the major show stopper for a full-on ARM onslaught into the desktop space is that "legacy" x86/x64 apps - which the whole world runs right now - either won't run at all, or will run poorly under some sort of emulator. The fact that all current ARM chips are physically optimised for low power rather than all-out data processing performance really doesn't help the situation, either.
For the moment, let's assume that this problem is successfully overcome, perhaps by porting various key apps over to ARM say. Due to the significant efficiency and performance improvements of the ARM architecture (see video below) x86 begins to be phased out, eventually disappearing. Now, where does this leave Intel? To go bust, obviously, as it can't sell any more x86 chips. No, of course not. Intel has had an ARM licence for years, so it seems logical that it would put its many superb data processing enhancement technologies into ARM chips, to create monsters that are capable of the blistering speeds we see today from x86 chips and then some. Intel really, really won't like this situation though. Why? Because at the moment, it's only proper competitor in the x86/x64 space is AMD, which conveniently for Intel, is some considerable way behind with its flagship Bulldozer architecture. One competitor. Easy to take care of. Could probably kill its x86 business if it wanted to, just by accelerating the performance of its chips by 50%. But then that pesky Competition Commission would start investigating…
However, the ARM CPU is made by literally hundreds of different companies, since ARM Holdings is a fabless company and makes its money by licensing the rights to make the processor. It doesn't take much of a stretch to see that some big hitter like IBM, who has similar expertise in building high performance processors (think PowerPC and POWER) could start competing with a high performance desktop variant of the ARM architecture. AMD will likely do the same, if they want to remain as a CPU manufacturer (they'd still have the profitable graphics card business to fall back on, so wouldn't die). Suddenly, Intel has lots of stiff competition from all sides and that extremely profitable niche that it has sat in for the last 30+ years due to licence exclusivity evaporates, perhaps eventually becoming a me-too commodity player with razor thin margins. Very painful, very humiliating, totally unthinkable. Maybe this is the real reason why Intel only ever made a half-hearted attempt with its Xscale ARM processors and the product line never really took off? It would have to literally be forced like this to make anything more of it.
So, you can see how it's completely in Microsoft's interest to move to ARM and absolutely not for Intel to do so. They now both want diametrically different things out of their long term relationship, so no wonder it's cooling off. Any bets on when the divorce papers will hit?
And now for that video. The short video below, originally found in an interesting geek.com article, compares a 1.6 GHz dual core Atom CPU in a netbook against a development board using a Cortex-A9 ARM CPU, configured as a dual core system, running at a mere 500 MHz. Yes, just 500 MHz. The results? Even with the netbook having a graphics accelerator and the ARM dev system not having one, the ARM was only slightly slower than the Atom! Of course, it consumed a lot less power than the Atom CPU too, which is critical. Note that this video dates from Jan 2010 and there's newer versions of both products now. However, it's still valid today, as the performance balance hasn't changed much between the two processor architectures. This is because the differences are inherent to them (x86 is hot and inefficient, basically) so it doesn't really matter how much each one is tweaked, the performance ratios will stay roughly the same.
84 Comments on Wintel Alliance Slowly Crumbling, ARM To Eventually Rule The Desktop?
ARM doesn't have much hope of competing on desktops. All the applications you currently use won't work (including all games previously made).
i guess it will be like that. ARM strong enough in smart phone market, they wont risk their future with battling in desktop/laptop against intel or amd.
edit: if they plan to get in maybe they gonna put in low or mid range, i think its pretty reasonable since they have power in it
We use x86 on the desktop because they are more powerful, not because they are the most power efficient.
The x86 architecture isn't going anywhere in the desktop market. Look how long it took for x64 to take over.
And lets talk about performance. The video only shows me one thing, how slow ARM is. I guess it is fine if you are rendering a few low end sites with no moving flash/HTML5 elements. And that is generally the site type accessed by smartphones. However, even rendering generic sites with just text and some graphics the ARM processor was noticeably slower. I'd hate to see it handle some flash sites, or even youtube with HD video running. I bet it would fall on it's face compared to the Atom.
Is there anyone here that honestly thinks Intel couldn't created an x86 processor that competes with ARM processors in performance and power consumption?
Look at what we already have. The Cortex-A9 has a TDP of 1.9w @2Ghz, the Atom Z550@2GHz is only at 2.5w. That isn't a huge margin for Intel to make up. Hell, Intel could probably make that up by cutting out the Hyperthreading silicon, dropping a few instruction sets and the silicon related to those, and cutting the L2 in half.
Or better yet, carve out a new product using the things they've learned with Sandybridge. I don't have a doubt they could manage very close to 2GHz with the same or better performance to what ARM can offer.
they made the ATOM chip, by castrating existing desktop architectures (cutting cache, pipelines, atc) and still they couldn't get a chip to perform for use in a smartphone or tablet that would sip less than a WATT of power under load, and less than 1/4 WATT at idle.
there process advantage does not make them magically have an architecture that performs efficiently in netbooks, tablets and smartphones. they tried for years, and only now they have a decent processor offering approaching tablet form with good battery life. they still have years to go before they enter the smartphone market.
ARM is encroaching on x86 in a bif way, because it's easier for them to evolve their architecture for more performance at the desktop and enterprise/server than it is for Intel to figure out how to make portable efficient chips for smartphones, etc.
with Microsoft making Windows on ARM, a whole new generation of apps and games will be made, that means market share AWAY from x86. the more developers that make ARM software, the less that x86 dominates.
you think that just because x86 is massive it is too big to lose it's crown?
you would do well to read up on some history, as well as notice what is happening in the portable OS and device market right now. would you have said Apple's iOS would always dominate because it had the most developers and software ? or would you realise that Android would take everything from Apple's mobile market and dominate around the world ?
things change, and right now there are HUGE changes happening in the OS and processor architecture worlds. the last 30 years of x86 are already eroding, and the same is happening for the desktop and mobile OS shares.
wintel is dying quicker than ever.
Now look at the current Sandybridge and what it is capable of. They have dual-core Sandybridge processors, full powered full featured chips, that sip just 17w. Image what they could get that down to just by ditching the silicon for HT, and the 3MB of L3 cache? And that is with a graphics processor, drop that and what does the power consumption numbers drop to? Cripple the memory controller(who needs dual channel 32GB RAM on a smart phone?), cut out the PCI-E lanes(don't need those on a smart phone), drop the unused instruction sets and the silicon related to those. What would the power consumption numbers look like then?
If Intel was worried, and they wanted to put their effort into it, I have no doubt they could put out an x86 processor today based on an extremely modified Sandybridge that could pull under 1w under load, and idle at under 0.25w and outperform the competition.
Guess what - it doesn't matter.
I can Skype, Facebook, send emails and play Angry Birds with a small fraction of the power already available today.
The only relevant 'tech specs' are that Intel is currently incapable of competing with ARM on the all-important idle power consumption. If they are able to address that then they can enter the smartphone race (albeit very late to the party). Intel is currently working very hard at this.
And to Wile E, people use Wintel because that's what the market determined was popular from the 1980s. They didn't choose it because of any tech related specification or power consumption.
Dorks bought a 386sx33mhz w/ 2mb of ram, a 80mb HDD and a Trident video card sitting on an ISA bus. People bought IBM-compatible PCs.
I like ARM because it's a comeptition and as such it can only be good for users but sometimes it's better to say nothing than talk BS...
Doing the same based on Sandybridge, where everything except for data I/O is on the die already, and they still manage 17w, gives a good example of what Intel can do. And if they put their minds to it, and see ARM as a serious threat, there is no doubt in my mind they could put something out using a highly modified Sandybridge that consumes 1W or less and idles at a fraction of a W.
But either way you look at it, they are still trying to compare 2 products that aren't meant to be compared. Atom is a full fledged processor for a laptop/desktop computer, while ARM is really a smartphone processor. And if ARM every does make a serious move to desktops/laptop, they will be nothing more than web/media devices at best, certainly not something that will rule the market.
"The Cortex-A9 has a TDP of 1.9w @2Ghz, the Atom Z550@2GHz is only at 2.5w."
This means that atom would be trashed by ARM performace wise, GPU is already way faster, not to talk about rest of SoC, oh, yeah, SoC, no slow buses etc.
Intel should however invest heavily in a serious answer to arm.... things change. Just ask the dinosaurs.
Intel's answer to ARM are Ultrabooks. Whether its their final answer, or the beginning of a series of answers I do not know. Either way their server business is still doing great, and I forsee them being the next IBM style company in a few more years(or decades).
Windows NT was available for Alpha and Sparc, Windows XP added x64 extensions even when it was primarily AMD's domain and Intel didn't have a 64bit x86 proc.
With ARM increasingly overlapping with netbooks and tablet PCs, it only makes sense for the next Windows version to support ARM.