I did an article on ARM and x86 about a year ago, since multicore ARMs were up and coming, and Windows 8 would allow them to be used in netbooks, notebooks or even desktops, as competing solutions to x86. That was how I saw things at the time: ARM moving towards x86.
But Intel has recently introduced a new generation of Atom processors, codenamed Medfield. This is showing a move in the opposite direction: x86 moving towards tablets and smartphones. The move in itself is not new, as Intel has been trying to get Atom processors in embedded and mobile devices for a while now. But the difference is that Intel is actually succeeding this time.
As I mentioned last time, x86 bears some legacy which makes it inherently larger and more complex than a more modern architecture, such as ARM. However, this extra overhead is more or less a constant factor, where the rest of the CPU design will grow larger over time as more transistors can be fitted on a single die, because of manufacturing progress. For regular desktop, workstation and server systems, the x86 overhead has become a non-issue years ago. With the amount of execution units, caches and everything you find in a modern CPU, the legacy overhead of the x86 becomes insignificant. An added factor is that Intel has always had the most advanced manufacturing. This allowed them to fit more transistors in a smaller space, using less power, which could somewhat compensate for the extra cost of x86 legacy.
For Atom however, the small scale of things meant that Intel could not quite get the upper hand over the competition yet. An important issue with early Atoms was that it consisted of the CPU and a separate chipset, where the competing ARM solutions were a System-on-a-Chip (SoC). As a result, Atoms were not small and energy-efficient enough for mobile devices such as tablets, let alone smartphones.
Then Intel introduced Oak Trail, the first SoC version of Atom. This makes it more compact and more power-efficient. It was still too powerhungry for phones, but tablets were more or less within Intel’s reach now. But since ARM solutions still had better performance and battery life, Atom-powered tablets never took off.
The new Intel Medfield is an SoC as well, but it looks a lot better than Oak Trail. Oak Trail was built on a 45nm process, but Medfield uses a 32nm process. This puts Intel in its favourite position: a step ahead of the competition, who are still on 40nm process. The result is that x86 can now hide its legacy very well. And this time we will actually be seeing this Atom SoC in a number of tablets and smartphones.
So where I was expecting ARM to invade Intel’s turf, Intel is now doing the opposite. I was not expecting Intel to close the gap just yet. But it seems that they are well on their way. And 22nm is just around the corner for Intel, which may allow Atom to take yet another step forward compared to ARM.
This newfound competition is also interesting for ARM itself. Because Intel throws HyperThreading into the mix, which ARM does not have. And Intel’s SIMD extensions may also be slightly more mature than ARM’s. So we will have to see how ARM is going to respond to this.
Scali's OpenBlog™ 
I have seen Intel grow its business at the exoense of the compentition before. This is not new for them.
ARM does not need to respond. ARM is already there, it’s Intel trying to breathe on their butt. But I admit competition is good. Very good for Intel.
Ofcourse ARM is going to have to respond. Intel has just launched a competitive chip, and has signed some big deals with OEMs.
If ARM does not respond by offering faster, better chips, Intel will take over the market in no time.
Considering the intense competition on the mobile platform it is pretty much a moot point whether intel can bring out a sustainable chip but has more to do with if it can survive with lowered profit margins there. Evidently, it could hurt intel in the long haul.
Well, to me it seems obvious that Intel is better equipped for competition in lower profit margins than any other company. For example:
1) Intel develops its own architectures, so there is no extra overhead for licensing from a third party (like ARM).
2) Intel has its own fabs, so no overhead for third-party manufacturing.
3) Intels fabs are generally a node ahead of the competition, making high-volume manufacturing cheaper.
4) Unlike most competitors, Intel is not solely dependent on its mobile products. Intel has various other product lines, which can compensate for investments done for the mobile market.
So ‘hurting’ is all relative. Perhaps Intel won’t be as profitable as they are today, but no other chip manufacturer is anywhere remotely as profitable as Intel is anyway. As long as Intel can get enough volume in the mobile market, they’ll probably remain the most profitable chip manufacturer.
This is the same attitude of Sun Microsystems, which made them to sink and today nowhere in server market.
Look 5 years down the line…
Pingback: x86 vs ARM: Intel’s next move is Clover Trail | Scali's OpenBlog™
True enough and I concur with most points. And yet most of this depends on whether or not intel makes significant headway into smartphones/tablets market or not. That’s a significant draw back for intel having its own fabs since maintaining these would require that it establishes a hegemony there considering the stagnating pc market and all. Medfield is an impressive chip considering that it carries a somewhat anachronistic ISA and for a single core SOC the performance is certainly remarkable. Ostensibly, intel still has a significant R&D advantage.
ARM has already made foray into intel’s territory and shall soon venture into the server market. This should be a fairly interesting scenario since I would like to see how ARM would progress there.Moreover, Globalfoundries intends to have 14nm finfet in volme manufacturing in 2014, the same timescale as Intel has for introducing 14nm finfet manufacturing so if GLobafoundries does actually meet the stated timeframe(A little unlikely) it would altogether mitigate the process advantage that intel has conjuring up a bleak situation for the chip manufacturer if things and x86 in particular go as envisaged by globalfoundries.
I am pretty excited since some competition would be good en masse. It is darwin’s law at its finest
I completely disagree with your assertion that Intel needs success in the smartphone/tablet market to survive. Intel is WAY, WAY bigger than any ARM-based company out there. Also, even though the desktop market is ‘stagnating’ at this point (as in not growing, but not really shrinking either), it is still a huge and very profitable market, where Intel has probably more than 70% marketshare at this point.
Aside from that, Intel does a lot more than just desktop processors. They have their server/workstation lines of CPUs, they have their HPC expansion cards (Xeon Phi), they have their SSD market, networking equipment, etc.
Intel is a ridiculous amount larger than any of its competitors, and they could afford to shrink somewhat, close a few fabs, and still be way ahead of the pack (not that I think this will happen on short notice, I think what Intel has been doing recently, in moving to Atom, SSD and Xeon Phi, will counter the developments in the desktop market).
This size advantage could pay off in the smartphone/tablet market. Namely, most phone manufacturers currently use various different suppliers for various different phone models. This means they have chips from different vendors, with different pin layouts, different instructionsets, different onboard GPUs and other hardware… All this requires them to put extra effort into designing new hardware, and porting the OS and drivers to each phone.
Intel is large enough to supply huge numbers of chips to these manufacturers, with a lot less variations in layout and specs (much more like the desktop market, where low-end i3s through high-end i7s are drop-in replacements). Then phone manufacturers could become more like Apple, who only needs to support a handful of different configurations for their product line.
It is also seemingly inevitable that Intel’s SoCs will truly eclipse ARM-based solutions in terms of performance and power consumption soon (probably in early 2013, when the 22 nm SoCs are rolled out). All this will make Intel SoCs a very interesting option for smartphones and tablets.
As for the server market… It seems that Intel has pre-empted that already. I don’t think ARM will be very interesting for compute-intensive scenarios. Conventional x86 and Xeon Phi/nVidia Tesla will be far more attractive options there. ARM will mainly be interesting for low-power servers, but since Atom is now competing directly with ARM in terms of performance-per-watt, I don’t give ARM too much of a chance. The server world has already been using Intel chips for years, more specifically x86, so why change to ARM if there is no real advantage?
Intel is also bringing down power consumption on the Core-series further with Haswell and future architectures, so at some point in the future, Atom and Core will meet, leaving ARM no room in the market whatsoever.
As for GlobalFoundries matching Intel on 14 nm: Not much of a chance. Intel has announced they will build Xeons, Cores and Atom products on 14 nm in 2014. That is not just ‘volume manufacturing’, but actual complex CPU/GPU designs at 14 nm (they intend to start volume manufacturing in late 2013).
In general, ‘volume manufacturing’ tends to start out with simple NAND memory chips or such. I highly doubt that they will have actual Bulldozer CPUs (or whatever architecture they have by then) in 2014. AMD’s roadmaps do not indicate such at any rate. Recent news is that the 28nm Steamroller has slipped to 2014, so 14nm is unlikely to come out in 2014 as well. Besides, AMD will probably have a 20nm node between 28nm and 14nm (Excavator?). Going directly from 28nm to 14nm seems extremely unlikely.
I know that Global Foundries has a poor track record and I am also fairly certain that Intel could bring in a more than competitive chip to the market. I do not doubt the chipmakers prowess here considering their nearly limitless resources.However when I state that for “Intel to survive”- it is the business model that I am alluding to. The reason that ARM has been so successful in the smartphone market is because it follows a business model that is largely volume driven and consists of low margins per wafer.I don’t know if Intel would be willing to acquiesce to such a business model that would be detrimental to their long term goals.
Intel has investors to answer to and for their fabs to remain profitable they must drive the requisite volumes.This is a fact. Moreover while the extant pc market is still largely profitable;the overall sales are declining with the computers being relegated to a secondary role and smartphones and tablets becoming the future. It is thus a moot point to discuss whether Intel can bring in a competitive chip or not but it is more important to assess if Intel is willing to acquiesce to drastically reduced margins per wafer for it to be relevant in such a market.Then again if Intel could achieve something of a hegemony with tablets by way of relegating arm to the low priced ones then that should be good enough.
As for Servers it remains to be seen if the ARM/AMD plans take off. They are of course targeting micro-servers and low powered ones.I don’t think that AMD is going to use Bulldozer here since what they are attempting is find a consonance with Jaguar and ARM. I don’t know why AMD has to employ ARM in such a scenario since I fail to understand the think-tank if they are contemplating getting into more verticals.
Really, I don’t understand your concerns at all. They seem horribly far-fetched as I already said.
“Remaining profitable” is not exactly one of Intel’s immediate issues, as I already said.
I also get the distinct impression that you did not even read anything I said, or at least did not take the time to think through the implications of what I said, because you just recycle the same arguments, that were already countered.
Pingback: Windows and ARM: not over yet | Scali's OpenBlog™