ARM Co-founder speculates on the future

[cbn]

"ARM is stating that first silicon will ship in 2012 at 32/28nm"

x86 will be way beyond that two years from now

1. The Medfield atom will be a big improvement over Moorestown atom. (Medfield is a single 32nm chip....whereas Moorestown is a separate 45nm CPU with a 65nm PCH).

2. Intel could also release "out of order" atom sometime after 2012-2013 according to this Anandtech article.

3. Furthermore, we don't know what kind of low power/mobile/3D architectures Intel is experimenting with in it's private fabs. Surely Intel is building something (using its massive R&D budgets) to benefit both its future mobile and future sever markets simultaneously.

Okay, now with that out of the way: I still think this idea of ARM moving out of Smartphone/Tablet area sounds really interesting.

 

[cbn]

this 2 GHz dual core A9 that will ship in a few months in China

Thanks for the video. (Interesting net-top idea)

 

[P4man]

NO THEIR DEPENDENCY ON WINDOWS HAS HELD THEM BACK. because they tied atom to windows so far people think of it as just an ultra low performance desktop platform, even though it is much higher performance than even the best arm chips. nobodys tried saddling up an atom with a slicker os like ios, android, palm or a lean linux

Oh common, intel themselves created moblin/meego for this very purpose, Joojoo already launched an Atom + custom linux tablet early this year. Its been canceled because they sold in the tens if rumours are to believed. Tens, not even tens of thousands. Finally dual boot android/windows tablets are already on the market, so dont tell me 'no one thought of it'.

The fact of the matter is, when you are going to make an android/chromeos/ubuntu/meego/linux device, arm simply offers the better solutions. Cheaper, smaller, better battery life and with DC cortex a9, competitive performance. Its only when you want to sell something that can run windows that atom makes any sense.

If you want proof of that; lets have a bet. HP bought Palm recently for webos, so they control the software ecosystem. They are intel's biggest partner and intel has them over a barrel with itanium if nothing else; will they port webos to x86 and sell atom based phones and (non windows) tablets? My money is on that they wont, no matter how hard intel twists their arm (pun intended); HP will pick the best architecture for the job or they will have bought Palm for nothing.

As for that rumored moorestown phone in 2011; Deja vue. Assuming these ones dont get canceled like all the previous atom phones, wake me up when intel grabs 5% of the mobile phone market. Actually, make that 1%.

 

[P4man]

"ARM is stating that first silicon will ship in 2012 at 32/28nm"

even if that were competitive clock for clock, x86 will be way beyond that two years from now

Clock for clock it will thoroughly trash atom. From whats been disclosed IPC will be a lot closer to pentium/athlon. Cortex A9 is competitive with atom clock for clock:

Though application benchmarking and making apple for apple comparisons in this segment is very difficult, a 1 GHz DC A9 is more than a match for a 1.6 GHz atom on a lot of tasks.

 

[Mark R]

Are you talking about phones with Cortex A8 vs Qualcomm Snapdragon "ARM" vs Marvell "ARM". I thought they all used the ARM instruction set?

The ARM v7 instruction set may be the same.

However, the extra features in the processors are very different between Marvell and Qualcomm. Different, and incompatible, vector instructions (Qualcomm use their own proprietary vector instruction set called VeNum, whereas Marvell use Intel's Wireless MMX2 instructions). Similarly, there are different security extensions, etc.

Don't forget that these devices aren't just processors - they're entire integrated systems. So a different processor, may include different peripherals: e.g. Qualcomm include a OpenGL2/mobile Direct3D compatible GPU, with various 2D/HTML acceleration features, whereas Marvell don't (although they do include H.264 decode hardware).

 

[P4man]

Thanks for the video. (Interesting net-top idea)

Did you notice what he said around 7:30 about partnering with microsoft? If I understood that correctly he was hinting the "next generation" windows release would run on ARM, and he didnt mean CE/phone, but something more powerful. In other words, windows 8.

 

[cbn]

Did you notice what he said around 7:30 about partnering with microsoft? If I understood that correctly he was hinting the "next generation" windows release would run on ARM, and he didnt mean CE/phone, but something more powerful. In other words, windows 8.

Yes, I watched the whole video and I did hear him mention Windows 8.

I guess it makes a degree of sense (to my layman perspective) after reading this Daily Tech article.

I also noticed he was quite interested in Chrome OS (but not necessarily Meego...at first he said yes to Meego, but immediately corrected that answer to No)

P.S. It also sounds like the finalized 2 Ghz Cortex A9 will not need a fan. (Apparently power consumption will drop from 2 watts on the sample to 1.5 watts on the production part)

 

[cbn]

I have a couple of questions regarding the 2GHz Cortex A9:

Are those chips capable of having RAM memory directly on the package? Or are they limited to RAM external to the chip for various reasons?

(In the video provided by P4 man, the 2Ghz Cortex A9 featured a memory DIMM)

I tried looking for an answer on the ARM website under Cortex A9, but couldn't find any notes for that.

 

[cbn]

The ARM v7 instruction set may be the same.
However, the extra features in the processors are very different between Marvell and Qualcomm. Different, and incompatible, vector instructions (Qualcomm use their own proprietary vector instruction set called VeNum, whereas Marvell use Intel's Wireless MMX2 instructions). Similarly, there are different security extensions, etc.

Isn't the Marvell processor found almost exclusively in Blackberry phones, whereas Qualcomm Snapdragon is found mixed in with a lot of Cortex A8 Android phones?

If that is true, it makes me wonder what Qualcomm's Android market share is compared to Samsung Humingbird, TI OMAP, Nvidia Tegra, etc. I mean, who would program for one type of processor if the resulting programs were incompatible with the majority of phones out there?

Don't forget that these devices aren't just processors - they're entire integrated systems. So a different processor, may include different peripherals: e.g. Qualcomm include a OpenGL2/mobile Direct3D compatible GPU, with various 2D/HTML acceleration features, whereas Marvell don't (although they do include H.264 decode hardware).

Speaking of GPUs, I am counting six so far for these smartphone/Tablet SOCs:

1. Mali (ARM's GPU)
2. Imagination Technologies
3. Adreno (Qualcomm's GPU)
4. Nvidia
5. Vivante
6. PICA

 

[P4man]

Yes, I watched the whole video and I did hear him mention Windows 8.

I guess it makes a degree of sense (to my layman perspective) after reading this Daily Tech article.

I dont see much of a direct relation. Virtiualization has nothing to do with binary compatibility. This wont make x86 code run on an ARM (or vice versa).

I also noticed he was quite interested in Chrome OS (but not necessarily Meego...at first he said yes to Meego, but immediately corrected that answer to No)

Since nufront seems to be aiming for desktop computing initially , that makes complete sense. Chrome also has a few features that might ease a transistion from wintel to arm, namely the ability to run apps remotely (on your pc) and stream them to your chromeos device.

P.S. It also sounds like the finalized 2 Ghz Cortex A9 will not need a fan. (Apparently power consumption will drop from 2 watts on the sample to 1.5 watts on the production part)

There is no such thing as "the 2ghz cortex". There will be this nufront chip, and there will likely be others. I kind of doubt you need active cooling for 2W though, 2W is about what a desktop memory module uses. He did say typical load playing movies, which is probably rather far from the TDP (im assuming this chip has a dedicated video decode unit).

If nufront implemented the TSMC 40G ARM hardmacro (which I strongly suspect, given the fact this is a startup ), then ARM say the core will use 1.9W @ 2 GHz. Add to that all the IO, the GPU and what not, and dont expect this chip to have a 1.5W TDP.

I have a couple of questions regarding the 2GHz Cortex A9:

Are those chips capable of having RAM memory directly on the package? Or are they limited to RAM external to the chip for various reasons?
(In the video provided by P4 man, the 2Ghz Cortex A9 featured a memory DIMM)

I tried looking for an answer on the ARM website under Cortex A9, but couldn't find any notes for that.

Not all ARM chips are equal. ARM provides the core IP and lots of optional modules (cache, gpu, neon, etc), Its the qualcoms, nufronts and nvidia's that mix and match and do the rest. Some implement die stacked ram (like the apple A4) for power/cost/size reasons, others work with external RAM. There is nothing inherent about ARM that prevents or necessitates this.

I mean, who would program for one type of processor if the resulting programs were incompatible with the majority of phones out there?

Its not unlike x86, where some chips may have SSEx, AVX or whatever, and others dont. In the old days some might have an FPU others not. The situation is even more comparable on the GPU side where you have several options and not all support the same feature set, though they will usually suppport a higher level API like DirectX or opengl.

Speaking of GPUs, I am counting five so far for these smartphone/Tablet SOCs:
dd
1. Mali (ARM's GPU)
2. Imagination Technologies
3. Adreno (Qualcomm's GPU)
4. Nvidia
5. Marvell

Those are the most important ones, certainly not the only ones. For instance you missed the PICA gpu used in the nintendo DS3. Your list is also slightly misleading. Most ARM vendors license GPU IP, just like they license the core IP. Marvell's GPU are based on IP from Vivante. Adreno and nvidia are the only proprietary ones Im aware off (and Adreno is actually purchased from ATI which used to license it).

 

[Mothergoose729]

There is a lot of assumptions there. Right now ARM has found a natural home in low power, low performance phones. The demand for more processing power in small devices is increasing, however the demand for more processing power in servers and desktops is not. Bottom line is there will never be a circumstance where any savings ARM can offer in power will mitigate the cost of transitioning to a completely different architecture without familiar software support. Even if ARM managed to offer more performance... there is just way to much history and way to much momentum behind x86. There have been plenty of cases where a standard has beaten out efficiency, and I think that is exactly what is going to happen here. ARM will have a very hard time competing against intel if moorestown achieves similar levels of efficiency.

 

[P4man]

There is a lot of assumptions there. Right now ARM has found a natural home in low power, low performance phones. The demand for more processing power in small devices is increasing, however the demand for more processing power in servers and desktops is not.

What?

Bottom line is there will never be a circumstance where any savings ARM can offer in power will mitigate the cost of transitioning to a completely different architecture without familiar software support.

Thats a sweeping statement, and IMO, an incorrect one on several levels.

First, the "familiar software support" is a bit of a canard.
In consumer space, to hundreds of millions of customers who own a mobile phone but no PC, their mobile phone OS and their browser is more familiar to them than windows. They would find it much and much easier to use a chromeos or android machine than a windows one. Even many of those somewhat familiar with windows will appreciate the speed, ease of use, security and simple maintenance. For a significant portion of the market, this is not a handicap, but a selling point, like OS-X is for Apple.

In the server space, linux rules. And for many applications, it just doesnt matter much what ISA that runs on. You simple need your LAMP stack, bind or whatever.

As for the power savings not mitigating the cost (if any); I beg to differ. In both datacenters and the mobile space, performance/W is the key selling point. If google/ms/facebook/oracle whomever can achieve a 50% power consumption reduction in their datacenters, that will certainly offset costs of porting a lot of software (that often wont even need to be ported at all).

Finally, if you offer an end user a choice between a mobile phone thats plenty snappy and with 2 day battery life or one which is faster in benchmarks but has less than a 1 day battery life, its not going to be a hard choice for most. Once its fast enough, its fast enough, but more battery life is always relevant.

Oh and did I mention cost yet?

Even if ARM managed to offer more performance... there is just way to much history and way to much momentum behind x86.

x86 isnt going away any time soon, if ever. Neither does wintel. But Im seeing this as quite similar to what happened in the 80s and 90s where our PCs were 5x or 10x as expensive, called workstations and ran Unix on a Risc CPU with SCSI disks. And no one believed we'd ever use a lowly intel cpu with IDE drives and a crappy consumer OS like windows. And yet, thats exactly what happened. Not because x86 was faster, nor was windows better, but high volume and low prices slowly killed the risc workstations nonetheless, from the bottom up, from PCs to workstations to servers, and that despite the performance handicap which was only overcome after it had become the defacto standard.

I see our x86 wintel PCs as the next generation of workstations. Many people will swear by it and never touch an ARM + chromeos pc or tablet or laptop if they dont have to. But the volume, price difference and ease of use will make that portion of the market increasingly a nerdy highend niche. The masses will buy the cheaper, more appliance like computing devices, and the software will follow the masses, just like it did for our workstations.

ARM will have a very hard time competing against intel if moorestown achieves similar levels of efficiency.

Thats a mighty big if, but even if, that doesnt make it a threat to ARM. What is Atom's advantage over ARM assuming comparable power and performance? That it can run windows desktop software. I dont see a big market for that in the mobile phone market, or even tablets for that matter. I think the opposite is true, Atom has a serious software handicap, all the mobile OS's and software stacks run on ARM primarily. As an illustration, Android for Atom is still stuck at version 1.6. How many of the 100.000 android market apps will run on x86? Its the same chicken and egg problem that you think will prevent ARM from competing in the x86 space.

But thats not even the biggest issue; a single (dual with some good will) monopolistic supplier who has to protect its mainstream product margins and cant afford a flexible way for customers to license IP and customize the silicon to meet specific needs, that will never succeed in these commodity markets. Intel has been selling in the embedded markets for decades, from i386 onwards, and their market share is not increasing as some people think with atom, its dwindling year after year. ARM has all but taken it over already. Its the business model that just doesnt work as well.

 

[sxr7171]

Today I was reading one of the older Daily Tech Chrome OS articles here, and if you scroll down to the 5th and 6th entries you will see the following comments on the matter:





The main question in my mind is what does Google ultimately want to accomplish with Chrome OS? Obviously selling user information is one of their major goals ("caveat emptor"/"buyer beware"), but does providing more offline apps (in addition to the offline media player that already exists for Chrome OS) help or hurt this?

Apparently the 3G purchase UI is one factor holding back the Chrome OS release, but does purchasing an expensive data plan for a cheap Chrome Book (that helps Google sell location based information in addition to its internet search information) really make any sense to frugal consumers likely to consider these products?

Maybe Google will compromise and see the need for releasing x86 and ARM app stores?

Opinions?

I don't know if 3G is necessary. You could use Wi-Fi at home, in the US carriers are marketing these tablets because they subsidize things to make the perceived cost look smaller to the average American consumer. After all you can't get any real traction in the US if you release a phone and say it costs $799. American consumers prefer, as the current state of affairs would suggest, to pay $299 and get gouged by a carrier for 2 years.

I don't think a Galaxy Tab or an iPad could replace a laptop for me yet, but it's pretty surprising how those devices cover about 85% of what you do with a computer even today. If ARM hardware really is getting as fast as they say it is then with optimized software it could be sufficient many people's computing needs. Yesterday I suppose people had 3-4 PCs in the house. Today they maybe have 2 PCs and 2-3 netbooks. Tomorrow it could very well be 1-2 PCs and 4-5 Tablets.

If you can browse, consume your media and send e-mail, you've covered what 90% of consumers need in a computer. They already can do those things, and in the future there's no doubt they'll do more.

As for Google I completely agree that I don't really like their intentions. But most people unfortunately don't care.

EDIT: BTW, I think Meego is Intel's bid to keep relevant to the way people are doing their computing these days.

 

[P4man]

EDIT: BTW, I think Meego is Intel's bid to keep relevant to the way people are doing their computing these days.

Moblin was an attempt to bootstrap Atom in the mid/mobile space. To compete with ARM, intel needed a better OS than windows CE and a much better suited OS than windows 7 (and be able to ditch all the windows legacy requirements like bios, PCI etc). It wasnt a bad idea, though Ive never been particularly impressed with the ubuntu moblin remix builds I tried, it certainly had (and assume has) some potential.

But now that its been merged with Nokia's maemo in to meego, ported to ARM and managed by the linux foundation, I think it may well end up being a serious shot in the foot; if meego ever takes off, it will help ARM as much, if not more than intel, since Nokia's huge ARM customer base pretty much guarantees meego will be more ARM than x86 centric and to compete in the netbook space, ARM is perhaps more in need of a popular and more suitable OS than intel is.

Anyway, I dont think meego will get very far and neither will samsungs bada. Add to that list webos, no matter how good it may be. At least two of those three will need to join forces to stand any chance against the google rollercoasters. I hereby predict a HP/intel/Nokia MegoOS partnership in 2011 Either that, or BadaWeb

 

[Mothergoose729]

What?

Ah, seems out there huh? Actually its true. PC gamers are really the only consumers driving the performance desktop market, most people buy lower performance laptops for mobility, or if they do buy a desktop, its the mini ITX/micro atx low power machines. Core2 era or equivalent processor can adequately support idle photo or video editing. The majority of desktop and laptop users don't desire more powerful computers. Even games are quickly falling behind the capabilities of hardware.

In the server space it is even more cut and dry. Except for the occasional super computer, almost no clients are buying multisocket configurations anymore. A huge majority of servers are single socket, a smaller proportion have dual socket, and that is based of recent sale figures. Most active servers are more then 3-5 years old and the owners have no plans of upgrading in the near future. There are HPCs being built today sure, but they are representing a smaller and smaller proportion of sales. Cost of ownership is becoming a bigger factor then performance in a lot of servers.

What his means for ARM is that intel and AMD architectures are likely going to migrate towards greater efficiency rather then peak performance, and you can already see this starting to happen with AMD bulldozer and more advanced power tricks form intel. The main advantage of ARM is perhaps cost and low power, but when many servers can be content with one or a pair of 45 watt processors in the future, where does ARM have a place to compete? 20 watts in power saving will not mitigate the cost of using probably inferior or at least less supported ARM software.

 

[P4man]

Ah, seems out there huh? Actually its true. PC gamers are really the only consumers driving the performance desktop market, most people buy lower performance laptops for mobility, or if they do buy a desktop, its the mini ITX/micro atx low power machines. [..]
What his means for ARM is that intel and AMD architectures are likely going to migrate towards greater efficiency rather then peak performance,

Slow down lol. So you are saying performance is no longer a key selling point, but power efficiency (and you left out cost, but let me add that too) is the key for consumer devices? And somehow that would work against ARM?

You know I quite agree with your point that demand for (CPU) performance is leveling out on the desktop (and notebooks), especially with gaming moving to consoles and handhelds, so there is a clear and growing market for devices that are "fast enough' for the most common tasks. Atom's success in SFF is a clear illustration of that. But if we accept that, then we've just eliminated 50% of ARMs problems entering the x86 market. The other 50% is software, but Apple and google and (ironically) intel are doing a good job eliminating that as well.

Look at it this way, if you dont need faster performance, dont need windows, why would someone prefer an x86 box with chromeos over an ARM box with chromeos? Let me rephraze, why would nearly everyone prefer the more expensive x86 box (or x86 notebook with worse battery life) ? Clearly we agree ARM has quite some potential here, no?

However, there is more to this and Im not even convinced ARM wont catch x86 on sheer performance in the longer run. Single threaded performance scaling is all but dead for x86, and has been for the last ~ 10 years. IPC improvements are rather miniscule and clock speed is barely going up. ARM is still increasing its IPC and clockspeeds at a much faster pace, but thats due also to shifting market focus. They dont have endless potential here either. Whats increasing is core count. And software is (slowly, but surely) adapting.

At some point, single threaded CPU performance will only matter for a tiny set of apps, the others will either not be performance sensitive at all, run on GPU's or require as many hardware threads as possible. Once you are there, guess which ISA can throw more cores for a given die size and power envelope? Its entirely feasible for a lot of workloads, high core count ARM chips will be faster than x86 in 5 or 10 years because the ISA is inherently more power and size efficient, since it doesnt carry all the legacy baggage, doesnt need complex instruction decoding and was designed from the ground up to be efficient on those metrics.

In the server space it is even more cut and dry. Except for the occasional super computer, almost no clients are buying multisocket configurations anymore.

Companies buy new servers to consolidate their old servers on. Virtualization, you may have heard of it. Whether single of dual socket doesnt matter too much, what matters is how many many VMs you can run per watt, per U and lastly, per $. If a single socket server offers that, thats what they will buy.

The other trend is outsourcing of your datacenter. Putting it in the 'cloud', renting what you need from providers like amazon, oracle and microsoft.

Anyway, again though, you are simply making the case for ARM. You will not want one to run your MS exchange server(s) on them, at least not as long as MS doesnt port it, but for a lot of tasks, you will get better performance per watt, per U and certainly per $ and in most cases, software compatibility isnt nearly as big a problem as on the desktop.

Keep a close eye on what Marvell is doing here, and calxeda (formerly known as smoothstone) . Since I like making predictions, Ill predict here they will be bought by oracle next year to revitalize their sun sparc server business and Ill predict they will make serious inroads in the datacenter market (not hpc) in the next 5 years.

Cost of ownership is becoming a bigger factor then performance in a lot of servers.

I couldnt agree more. Well, except that they are related, as higher performance per unit means more users/vms per unit, lowering cost Power consumption is the number 1 factor in TCO for datacenters though. Thats why ARM is betting on this market with cortex a15, and they stand an excellent chance.

As for cost; how can intel fight back? Are they going to position $50 atoms against it, and against their own $500 xeons? There is a reason you cant have ECC on atom. That many atoms dont come with virtualization extensions. Some dont even support AMD64. Intel doesnt want atoms in the datacenter (despite many companies selling you just that). ARM and its licensees are not held back by any of this. They dont have a higher margin business to protect, they have an opportunity to upsell $10 socs with some slight changes in to a market where they can charge 20x as much and still be considered dirt cheap and more efficient.

Interesting times ahead

 

[sxr7171]

Moblin was an attempt to bootstrap Atom in the mid/mobile space. To compete with ARM, intel needed a better OS than windows CE and a much better suited OS than windows 7 (and be able to ditch all the windows legacy requirements like bios, PCI etc). It wasnt a bad idea, though Ive never been particularly impressed with the ubuntu moblin remix builds I tried, it certainly had (and assume has) some potential.

But now that its been merged with Nokia's maemo in to meego, ported to ARM and managed by the linux foundation, I think it may well end up being a serious shot in the foot; if meego ever takes off, it will help ARM as much, if not more than intel, since Nokia's huge ARM customer base pretty much guarantees meego will be more ARM than x86 centric and to compete in the netbook space, ARM is perhaps more in need of a popular and more suitable OS than intel is.

Anyway, I dont think meego will get very far and neither will samsungs bada. Add to that list webos, no matter how good it may be. At least two of those three will need to join forces to stand any chance against the google rollercoasters. I hereby predict a HP/intel/Nokia MegoOS partnership in 2011 Either that, or BadaWeb

HP/Intel/Nokia would be nice. I don't think any one of those is capable of going it alone.

 

[Mothergoose729]

Slow down lol. So you are saying performance is no longer a key selling point, but power efficiency (and you left out cost, but let me add that too) is the key for consumer devices? And somehow that would work against ARM?

You know I quite agree with your point that demand for (CPU) performance is leveling out on the desktop (and notebooks), especially with gaming moving to consoles and handhelds, so there is a clear and growing market for devices that are "fast enough' for the most common tasks. Atom's success in SFF is a clear illustration of that. But if we accept that, then we've just eliminated 50% of ARMs problems entering the x86 market. The other 50% is software, but Apple and google and (ironically) intel are doing a good job eliminating that as well.

Look at it this way, if you dont need faster performance, dont need windows, why would someone prefer an x86 box with chromeos over an ARM box with chromeos? Let me rephraze, why would nearly everyone prefer the more expensive x86 box (or x86 notebook with worse battery life) ? Clearly we agree ARM has quite some potential here, no?

However, there is more to this and Im not even convinced ARM wont catch x86 on sheer performance in the longer run. Single threaded performance scaling is all but dead for x86, and has been for the last ~ 10 years. IPC improvements are rather miniscule and clock speed is barely going up. ARM is still increasing its IPC and clockspeeds at a much faster pace, but thats due also to shifting market focus. They dont have endless potential here either. Whats increasing is core count. And software is (slowly, but surely) adapting.

At some point, single threaded CPU performance will only matter for a tiny set of apps, the others will either not be performance sensitive at all, run on GPU's or require as many hardware threads as possible. Once you are there, guess which ISA can throw more cores for a given die size and power envelope? Its entirely feasible for a lot of workloads, high core count ARM chips will be faster than x86 in 5 or 10 years because the ISA is inherently more power and size efficient, since it doesnt carry all the legacy baggage, doesnt need complex instruction decoding and was designed from the ground up to be efficient on those metrics.

Companies buy new servers to consolidate their old servers on. Virtualization, you may have heard of it. Whether single of dual socket doesnt matter too much, what matters is how many many VMs you can run per watt, per U and lastly, per $. If a single socket server offers that, thats what they will buy.

The other trend is outsourcing of your datacenter. Putting it in the 'cloud', renting what you need from providers like amazon, oracle and microsoft.

Anyway, again though, you are simply making the case for ARM. You will not want one to run your MS exchange server(s) on them, at least not as long as MS doesnt port it, but for a lot of tasks, you will get better performance per watt, per U and certainly per $ and in most cases, software compatibility isnt nearly as big a problem as on the desktop.

Keep a close eye on what Marvell is doing here, and calxeda (formerly known as smoothstone) . Since I like making predictions, Ill predict here they will be bought by oracle next year to revitalize their sun sparc server business and Ill predict they will make serious inroads in the datacenter market (not hpc) in the next 5 years.

I couldnt agree more. Well, except that they are related, as higher performance per unit means more users/vms per unit, lowering cost Power consumption is the number 1 factor in TCO for datacenters though. Thats why ARM is betting on this market with cortex a15, and they stand an excellent chance.

As for cost; how can intel fight back? Are they going to position $50 atoms against it, and against their own $500 xeons? There is a reason you cant have ECC on atom. That many atoms dont come with virtualization extensions. Some dont even support AMD64. Intel doesnt want atoms in the datacenter (despite many companies selling you just that). ARM and its licensees are not held back by any of this. They dont have a higher margin business to protect, they have an opportunity to upsell $10 socs with some slight changes in to a market where they can charge 20x as much and still be considered dirt cheap and more efficient.

Interesting times ahead

I will agree with you that ARM has the potential to offer desktops and servers similar levels of performance of intel chips, perhaps even greater performance in the future. And in terms of IPC IPW ARM can almost certainly match intel chips because they are not weighed down by x86 and associated legacy instructions. And absolutely, ARM can provide chips to servers for cheaper. The biggest problem is, in the world of servers, upfront cost is really not an issue. Honestly, most of the servers offered on Dell's or HP's website don't pack any more muscle then a desktop machine that is 1/4 the price, but they sell not because they are great deals, but because their is both hardware and software IT support that the OEM and others can provide. Reliability is a corner stone of servers, and standards are more reliable than something which is nonstandard. If the server market is going to migrate towards ARM, it will necessarily have to happen after consumer computers have already switched over to ARM. There will need to be a lot of momentum behind both software and developers, and face it, not a whole lot of people are coming out of grad school specializing in performance orientated ARM based computers.

As hard as it is to get servers to migrate to new things, it is even harder to get consumers to make big changes. Microsoft office, microsoft operating systems, adobe and microsoft video editing and photo editing tools, popular web browsers, remote desktop clients, GUIs, gaming APIs, literally every major piece of consumer and enterprise software has grown and developed around intel based computers. It takes more then a wave of a magic wand to get them to work with ARM processors, and it takes at least some incentive on the part of the developers to get them to even try. The first ARM smartbooks will probably use linux or some smart phone OS, which will help them sell as novelties like the ipad, but will do very little at all to displace desktop and laptop sales. Touch screen devices are great and trendy, but the peripherals associated with intel based computers will always be more practical for every day tasks. I can't imagine a scenario that would change that or cause any shift in that kind of thinking. I think if anything, people have shown time and time again how reluctant they are to completely abandon windows based computers. At many times in recent history windows has been an inferior operating system, or been coupled with inferior hardware, to other computers on the market. It all revolves around the circular logic "We use it because we like it, we like it because we use it".

Given time and a lack of competition, ARM could probably start to make major headway. The problem is that intel is eyeing the same market, and when moorestown based phones and tables are able to run a windows operating system it will be game over for ARM penetration. You mentioned how successful the netbook was being a clear example of "good enough" performance. Netbooks were failing pretty miserably until Asus and others started shipping them with windows operating systems. Wintel is a combination of two giants, and in whatever market the two meet each other in, everything else tends to fall by the way side.

 

[P4man]

Reliability is a corner stone of servers, and standards are more reliable than something which is nonstandard. If the server market is going to migrate towards ARM, it will necessarily have to happen after consumer computers have already switched over to ARM.

And why is that? Servers needed to be backed by a reputable OEM, but the notion that intel (or amd) chips would be more reliable than arm is just nonsense and isnt going deter any customer with a clue. The big customers dont even mind mind buying VIA servers from Dell if it lowers their powerconsumption:
http://en.community.dell.com/dell-b...no-based-server-for-hyperscale-customers.aspx

If you mean reliable in terms of RAS features, then yeah, those are lacking on ARM atm, but thats obviously a result of their current target market and Marvell and others are already fixing that. Besides, Atom is really a player in the server market (Im sure youve seen boxes such as this one http://www.seamicro.com/ ) and it doesnt even have ECC.

There will need to be a lot of momentum behind both software and developers, and face it, not a whole lot of people are coming out of grad school specializing in performance orientated ARM based computers.

First of all, almost no one develops in machine code anymore, so hardly any application developer is (or should be) really a specialist in either x86 or ARM or Power or whatever. Some knowledge obviously helps if you are doing really performance sensitive stuff, but by and large, its the compiler (or JVM) doing it for you.

Secondly, its been a while, but Ill tell you if I where still an ICT student now, Id very much make sure I learned ARM, linux/android, opengl. Thats where the money is right now. If you are a competent developer for mobile platforms, you get a job (though you probably wont take it, but write your own apps for the appstore and android market). If you are yet another microsoft certified MCAD, good luck.

As hard as it is to get servers to migrate to new things, it is even harder to get consumers to make big changes. Microsoft office, microsoft operating systems, adobe and microsoft video editing and photo editing tools, popular web browsers, remote desktop clients, GUIs, gaming APIs, literally every major piece of consumer and enterprise software has grown and developed around intel based computers. It takes more then a wave of a magic wand to get them to work with ARM processors, and it takes at least some incentive on the part of the developers to get them to even try.

Pretty much all of the above (if you remove brand names) is readily available for ARM. Have you tried ubuntu, it literally ships with all of the above, and the arm port makes available 95% of its vast repositories. Thats the beauty of opensource. It took canonical a week (and a rack full of machines) to recompile the repo's for ARM.

As for incentive for software developpers. Seriously. 120 million iOS devices sold to date, 300k applications, 7+ billion downloads. This is the healthiest software eco system out there. And Android is catching up with that rapidly. ChromeOS is to be launched still, as is meego and the others.

Most of those apps are obviously geared towards mobile phones, and increasingly, tablets, but thats more than a healthy base to jump towards netbooks and desktops. Especially if more desktop specific apps like videoediting, office suites etc are only a recompile away (if that).

The first ARM smartbooks will probably use linux or some smart phone OS, which will help them sell as novelties like the ipad, but will do very little at all to displace desktop and laptop sales.

Actually no, its not helping at all. Look at the toshiba ac100. Its a nice machine and Id love to have one, if only I didnt have to use android 2.1 on it. Thats simply not a suitable OS for it (yet). Its a great mobile phone OS, its an okay but still troublesome tablet OS, but at this point its still utterly rubbish as a notebook/pc OS.

But those are growing pains. OEMs are just short sighted here, they see android selling by the gazillions on phones and therefore use that, rather than a more sensible OS which they dont see sell that well in any market. If and when ubuntu matures on the ac100, Ill buy one in a heartbeat.

All it takes is one OEM to see the light and pick an OS and application stack that makes sense for such machine. Today that would be ubuntu IMO, tomorrow that might be android, meego, chromeos or whatever. Maybe even windows. There will be plenty of viable options.

Touch screen devices are great and trendy, but the peripherals associated with intel based computers will always be more practical for every day tasks.

Always is a very long time. Dont believe it.

I can't imagine a scenario that would change that or cause any shift in that kind of thinking.

But its already happening. In one or two years, the devices most often used to access the internet will be.. mobile phones. Those are the machines covering our basic computing needs. The cloud is handling our processing needs. Thats already happening (checked out Otoy or OnLive for instance?) Only the ergonomics suck, you need a bigger screen, and you have a tablet. And perhaps a keyboard, and you have smartbook or nettop.

I think if anything, people have shown time and time again how reluctant they are to completely abandon windows based computers. At many times in recent history windows has been an inferior operating system, or been coupled with inferior hardware, to other computers on the market. It all revolves around the circular logic "We use it because we like it, we like it because we use it".

Then explain to me why nearly no one is using windows mobile (anymore, it used to have like 30% of the smartphone market).

Given time and a lack of competition, ARM could probably start to make major headway. The problem is that intel is eyeing the same market, and when moorestown based phones and tables are able to run a windows operating system it will be game over for ARM penetration.

You really believe that? Honestly? First of all, moorestown doesnt even run windows (NT kernel, I dont assume you are referring to CE/mobile/phone which is struggeling not to become completely irrelevant), but who would even want to? Secondly, being able and being succesful are two very different things. If intel ever proves successful in the ultra portable space (and I have very serious doubts about that), it will be running the same OSs and same application stacks as ARM.

You mentioned how successful the netbook was being a clear example of "good enough" performance. Netbooks were failing pretty miserably until Asus and others started shipping them with windows operating systems.

No not really. Asus managed to cobble together one of the worst linux distro's ever seen for the initial eeepc, and it wasnt received well. If you have looked at it, that shouldnt surprise you. OEMs need to learn they are no good at OSs. Asus learned that with EeePC, but all of them are repeating that mistake now with slapping buggy crappy interfaces on top of Android. They will learn. Meanwhile Dell and HP and other seem to be doing just fine with ubuntu :
http://www.itproportal.com/2009/2/25/third-dell-inspiron-mini-9-netbooks-comes-ubuntu/

(Even if neither could resist the urge to implement a crappy custom shell on top of ubuntu).

Anyway, even if we assume 2/3 of netbook customers insist on windows for the moment, which I doubt, its probably more ignorance, but even then 1/3 of the market is pretty damn significant when you have zero today (talking about ARM netbooks).

Wintel is a combination of two giants, and in whatever market the two meet each other in, everything else tends to fall by the way side.

The wintel alliance is falling apart. ARM and linux are driving a wedge between them. Intel is sponsoring and pushing meego, its not even supporting windows with its latest mobile chips, while otoh microsoft seems to go and support ARM for its next generation desktop OS.

Intel can not succeed in the mobile space with windows, it had to push and support the alternatives, and microsoft can not afford to ignore the ARM gravvy train and let everyone run linux variants on them. I wrote this in the beginning of the year:

rcf wrote:
What will be Microsoft's reaction to ARM's competition with x86?
Just use Windows CE and be done with it?

Thats the $1M question.

The wintel duopoly is being challenged by ARM + Linux. Intel seems to respond to that by pushing x86 atom + Linux (see Moblin/Meego), I suspect MS is going to respond with a ARM + windows alternative, whether that will only be windows mobile (/phone) 7 or a port of windows 7 remains to be seen. But I cant see them do nothing, they lost the smartphone market not doing anything, I cant imagine they will allow to lose the tablet/netbook market without putting up a fight, because losing that market to android/chromeOS/meego/linux will provide a credible beachhead to arm+linux to attack windows' core markets.

Kind of fun to watch, both parties in the wintel duopoly pushing part of the arm+linux alternative

 

[cbn]

http://chromeossite.com/2010/11/25/75k-google-branded-‘speedbooks’-coming-soon/

This article mentions Chrome OS will initially use Intel then ARM chips (confirmed in this video @ 43:02)

 

[cbn]

Those are the most important ones, certainly not the only ones. For instance you missed the PICA gpu used in the nintendo DS3. Your list is also slightly misleading. Most ARM vendors license GPU IP, just like they license the core IP. Marvell's GPU are based on IP from Vivante. Adreno and nvidia are the only proprietary ones Im aware off (and Adreno is actually purchased from ATI which used to license it).

I'll Change Marvell to Vivante and add PICA to the GPU list. (Thanks for the corrections to this as well as my mistake of ARM and Windows 8. I also appreciated the information relating to ARM chip configuration, instructions, etc)

 

[P4man]

One more thing I forgot to mention; a few months ago Microsoft obtained an ARM architecture license:
http://www.eetimes.com/electronics-news/4204864/Microsoft-ARM-deal-game-changer

Very few companies have that. Most, like nVidia, Apple, and TI just license a specific core, like the cortex A8 or A9, but some want to make their own, different, but ARM compatible cores, like Samsung and now Microsoft, and that requires such architecture license (they dont come cheap).

That microsoft got one is extremely interesting. You dont buy one unless you intend to invest heavily in a custom CPU and custom core. Now I dont know what MS plans with this, your guess is as good as mine; it could be consoles (next xbox), it could be for mobile phones, mimic Apple's vertically integrated approach, servers or anything else. But clearly MS is taking ARM very very seriously.

So while intel is developing linux OS's to compete with windows, MS is building ARM cpu's (and obviously the software that will run on them) that will compete with intel chips. Take it from me: the wintel alliance is dead.

 

[cbn]

Touch screen devices are great and trendy, but the peripherals associated with intel based computers will always be more practical for every day tasks.

This has been addressed by both Google and Meego:

Google: Android for Touch, Chrome OS for keyboard and mouse.

Meego: Same core for all devices (coupled to a different UX depending on whether device uses Touch or Keyboad/mouse)

I also think it is interesting that Meego started off primarily as a Netbook Operating system. More information here (The handset UX didn't come until the later releases).

 

[grimpr]

One more thing I forgot to mention; a few months ago Microsoft obtained an ARM architecture license:
http://www.eetimes.com/electronics-news/4204864/Microsoft-ARM-deal-game-changer

Very few companies have that. Most, like nVidia, Apple, and TI just license a specific core, like the cortex A8 or A9, but some want to make their own, different, but ARM compatible cores, like Samsung and now Microsoft, and that requires such architecture license (they dont come cheap).

That microsoft got one is extremely interesting. You dont buy one unless you intend to invest heavily in a custom CPU and custom core. Now I dont know what MS plans with this, your guess is as good as mine; it could be consoles (next xbox), it could be for mobile phones, mimic Apple's vertically integrated approach, servers or anything else. But clearly MS is taking ARM very very seriously.

So while intel is developing linux OS's to compete with windows, MS is building ARM cpu's (and obviously the software that will run on them) that will compete with intel chips. Take it from me: the wintel alliance is dead.

I agree. It started with Marc Tremblay back in 2009.

http://www.theregister.co.uk/2009/04/08/sun_chip_geek_quits/
http://www.theregister.co.uk/2009/04/09/microsoft_tremblay_rock/

 

[cbn]

One more thing I forgot to mention; a few months ago Microsoft obtained an ARM architecture license:
http://www.eetimes.com/electronics-news/4204864/Microsoft-ARM-deal-game-changer

Very few companies have that. Most, like nVidia, Apple, and TI just license a specific core, like the cortex A8 or A9, but some want to make their own, different, but ARM compatible cores, like Samsung and now Microsoft, and that requires such architecture license (they dont come cheap).

That microsoft got one is extremely interesting. You dont buy one unless you intend to invest heavily in a custom CPU and custom core. Now I dont know what MS plans with this, your guess is as good as mine; it could be consoles (next xbox), it could be for mobile phones, mimic Apple's vertically integrated approach, servers or anything else. But clearly MS is taking ARM very very seriously.

So while intel is developing linux OS's to compete with windows, MS is building ARM cpu's (and obviously the software that will run on them) that will compete with intel chips. Take it from me: the wintel alliance is dead.

Thanks. That article had some great information.

I thought it was particularly interesting that Microsoft was looking for an engineer so it could run ARM in its bing data centers.

A previous job advertisement provides another clue. In April we reported that Microsoft was looking for a senior software development engineer to help with its Bing data centers, potentially running them on ARM hardware. That engineer is assigned to the Bing Autopilot Hardware team and the ad seemed to imply that an ARM processor could be used for the main server in an attempt to save power.

There was also speculation about MS having good reasons to support ARM on one of their future PC Operating systems.