backtop


Print 12 comment(s) - last by s12033722.. on Nov 5 at 6:30 PM


  (Source: ExtremeTech)
Decision is puzzling given Intel's potential to catch up in the tablet market

Intel Corp. (INTC) dropped a seeming bombshell [press release] this week when it announced that it is expanding its relationship as contract chip fabricator for a small, relatively unknown chipmaker Altera Corp. (ALTR).  

I. The Announcement -- Intel is Building ARM Chips (Some Restrictions May Apply!)

The announcement wasn't so much a surprise due to Intel looking to increasingly play third-party fab -- it's been known for some time that Intel's been exploring that new market role.  

What is more surprising (on the surface) is what Intel has agreed to produce: the newly announced tie-up will involve Intel fabricating at 14 nm Altera's 64-bit processors based on ARM Holdings Plc.'s (LON:ARM) proprietary architecture.  Given ARM and Intel's adversarial relationship, the move inevitably triggered a great deal of wild speculation, much of it focused on highly unlikely outcomes.

The move does raise some very intriguing, albeit speculative possibilities.  But I think it's important to carefully analyze where Intel stands in the mobile and traditional PC markets, before jumping to such conclusions.   Speculation, after all, is a fun game for investors and analysts alike, but it's wise to be principled in your speculation, at least, and understand the unknowns.  With that in mind let's look at why some shocking speculative scenarios are less likely and why others are slightly more likely -- if still far from a sure bet.

Altera FPGA
Intel will be making a chip with ARM cores onboard for Intel... but it won't be competing with Intel's products.  [Image Source: Wikimedia Commons]

To put this event in perspective, you must first recognize that ARM Holdings is Intel's top architecture competitor.  Intel has traditionally dominated compute sensitive server and PC chips, but has struggled to compete with ARM in mobile chips.  

ARM Inside
Yep -- that -- happened.  But maybe not in the way you're thinking. [Image Source: MaximumPC]

Intel wants x86 to conquer all computing spaces -- including mobile -- and is trying to leverage its process lead to make that happen.  However, it's been slowed by a lack of inclusion of 4G cellular modems on-die and difficulties adapting to the mobile market's low component prices.  ARM, meanwhile, wants a piece of the PC and server markets, but has received a lukewarm response from consumers due to software compatibility concerns.

The disappointing sales of (x86) tablet products using Microsoft Corp.'s (MSFT) Windows 8 and the flop of Windows RT (ARM) product in general somewhat unexpectedly had the net result of being a driver to maintain the status quo, allowing neither company to gain much ground.  For Intel, its partnership with Microsoft (the historic "Wintel" combo) has damaged its mobile efforts, as Windows 8 flopped in the tablet market.  Likewise ARM's efforts to score PC market share were stifled by the flop of Windows RT, which led to OEMs killing off ARM-based laptops and convertibles.

Haswell Chromebooks
Intel has moved away somewhat from Microsoft of late, eyeing other vehicles for mobile success.

Both companies seem to have learned their lesson and are migrating away from Windows towards other platforms -- in ARM's case Chromebooks, and in Intel's case Android tablets/smartphones.

But suffice it to say, ARM Holdings and Intel are still very much bitter enemies from a sales perspective.

And therein lies a major caveat: the new chips -- which Altera dubs the "Stratix 10" series -- will be used in high-end networking equipment and communication equipment.  Thus they won't be competing with Intel's Xeon server chips, Core Series PC processors, or Atom mobile chips.

The Altera/Intel chip will be a highly power efficient, highly flexible design incorporate both ARM Cortex-A53/A57 CPU cores, and FPGA reprogrammable circuitry onboard.  The ARM cores will handle fundamental computing needs, while the FPGA circuitry can be tailored for complex custom tasks such as digital signal processing.

ARM Cortex A53/A57
The Altera DSPs may be the first quad-core 64-bit ARM chips. [Image Source: ARM]

Apple, Inc. (AAPL) made waves in September when it announced its A7, a 64-bit chip based on ARM Holdings' A64 architecture and ARMv8 instruction set.  The A7 powers the iPhone 5S, Apple's flagship smartphone, which launched last month.

ARM is pushing partners to adopt 64-bit early to build up a catalog of apps, compilers, and expertise, in preparation for a push into the lucrative server market.  In order to convince its partners to make that move, it's pushed a number of improvements exclusively into its 64-bit instruction set and intellectual property cores, including the inclusion of more registers and better parallel computing capabilities.  

In addition to the obvious, but lesser justification that adopting 64-bit is "future prepping" for when smartphones eventually have more than 4 GB of memory, ARM's efforts to incentivize the switch appear to be a key reason why Apple chose to go 64-bit despite not currently needing to do so from a memory addressing standpoint.

II. The Common Line And Falsely Tempting Line of Subsequent Speculation

Ironically, the Altera-design, Intel-produced FPGA/processor combo may be the first quad-core 64-bit chip (Apple's A7 is dual-core).  But the chip doesn't do much to help ARM's software compatibility push, as it likely won't have much in the way of general purpose enterprise software written for it, given its specialist use.

It is of course worth noting that this is not Intel's first dance with ARM.  In the 1990s Intel produced the "StrongARM" processors -- often found in Windows "Pocket PCs", in many ways the predecessor of the modern smartphone.  Intel acquired the StrongARM unit via purchasing it from Digital Equipment Corp. (DEC) in 1997.  In 2000 it transitioned to a new family of ARM CPUs called XScale, which it would produce until it sold the unit to The Marvell Group Ltd. (MRVL), a French-owned chip design company.

StrongARM
In the 90s Intel produce ARM mobile processors. [Image Source: Chris's Acorns]

Since then Intel has stuck exclusively to the x86 architecture for consumer CPUs, although its enterprise offerings include a low volume product using Intel's alternative Itanium architecture, as well.

Given that history it's inevitable that some may read something into this announcement as it marks the first in nearly a decade that Intel will build ARM CPUs in its fabs.  Nathan Brookhaven, principle analyst at Insight 64, comments to Forbes:

It’s huge. Imagine ARM’s most powerful and technologically advanced 64-bits processor built on Intel’s leading-edge fabs. A duo that will be hard to beat.  Intel will build Apple's A7, Qualcomm’s Snapdragon or the Nvidia Tegra for the right price. Now, the question is, are they ready to pay that premium and feed their direct competitor, except for Apple. But that would actually make business sense for everyone.

I believe there's very little chance this is accurate, and even if the prediction does come true, it's not for the reason you're thinking.

III. Profit vs. Risk -- Understanding the Modern CPU Food Chain

Let me explain.

Whether it's tablets or PCs, the processor is still one of the most expensive components onboard.  Aside from the discrete GPU -- if a device has one -- the CPU has the greatest earning potential for a large company like Intel because the CPU is the most complex component.

Other components like the power supply or memory tend to either be lower margin or have more competitors.  The display, memory, and storage components are all sensitive to process, but see profit split between different parties (e.g. the company who makes the DRAM chips and the company who sells the stick of DRAM) and are primarily dependent on process technology.

CPUs and GPUs remain the toughest product to make, as it's not enough to simply have the best process, you must also have the best architecture and the best optimization of that architecture for the space you're competing in.

There's essentially five points of potential profit on the processor food chain:
  1. [CPU] Fabrication
  2. [CPU] Architecture design
  3. [CPU] Optimization
  4. OEM
  5. OS platform
Of these, the fabrication/OS point is the most profitable (but is dependent on the number of OEM adopters).  The second most profitable niche is optimization (which again is dependent on OEM adopter market share), followed by OEM markups.  In terms of expense, fabrication and operating system designs requires the greatest capital investment and the highest risk.
  

Fabrication
Fabrication is high risk and low margin.  Hence for companies like Samsung and Intel's it's only desirable when used as a vehicle to score sales in more profitable parts of the product food chain.

In terms of difficulty/risk, the fabrication and operating system are the most difficult/risky points.  Hence in terms of combined risk, cost, and profitability the ranking of which points are "best" is arguably:
  1. Optimization
  2. Architecture design
  3. OS platfrom
  4. OEM
  5. Fabrication
...with the fabrication point being last largely because it's so high risk.

In other words, the last thing Intel wants is to settle into a niche of playing fabs for everybody else's product, as that's an unsound approach.  If you can't keep up in terms of chip design, you typically spin off your fabs and opt for a different architecture direction -- just look at Advanced Micro Devices, Inc.'s (AMDspinoff of GlobalFoundries and upcoming ARM product to see that.

IV. Top Firms' Role on That Food Chain

Apple has seen unbelievable profits due to this fundamental premise.  It controls the two most desirable points on the food chain -- OS and optimization -- while sharing some profit with its architecture designer (ARM Holdings) and a bit with the fabricator (Samsung Electronics Comp., Ltd. (KSC:005930)).  By choosing to play operating system maker, too, it adds to its profits, but also its risk.  Note that nearly every other first-party exclusive smartphone platform has failed or is about to fail (i.e. BlackBerry, Ltd. (TSE:BB) and the now-dead Palm).

Intel controls points 1, 2, and 5, currently, on the food chain.  Compared to Apple, Intel's points of control offer less risk, but also slightly less profitability. Its architecture control may be at risk, but even so, it's currently the top in its most risky/expensive point of control (fabrication), where as Apple's most risky/expensive point of control (OS development) is much less of a clear leader (as Android has surpassed Apple in market share).  Hence Apple might be a better short-term investment, but Intel certainly appears a better long-term investment.

22 nm Atom
Architectural optimization is what truly produces a profitable product -- and oft overlooked point.  [Image Source: Jason Mick/DailyTech LLC]

Samsung is another top company in terms of market dominance and profit.  It occupies points 1, 3, 4, and 5 -- sometimes.  Sometimes Samsung's devices use third-party optimization firms like Qualcomm Inc. (QCOM) and NVIDIA Corp. (NVDA), which hurts profitability by removing one of the most profitable roles.  But Samsung makes up for this by being one of the largest and most successful third party manufacturers.

Microsoft enjoys a lot of profit due to its OS dominance, as does Google Inc. (GOOG); but both companies are limited in controlling only one point which they monetize in different ways (Microsoft by direct sales; Google by giving away OS product for free in return for web services market share and by proxy search advertising revenue).

Adreno inside
Qualcomm pulls in impressive earnings solely by optimizing ARM Holdings' architecture.

Qualcomm and NVIDIA are also quite profitable operating solely as optimizers, as is ARM Holdings who serves as architecture maker to Qualcomm, NVIDIA, Apple, and Samsung.

V. Four Scenarios in the x86 vs. ARM Competition

With that in mind let's assume four scenarios.  

Scenario one is that x86 proves dominant in the mobile space, assuming a comparable process.

In this case Intel has nothing to lose as serving as fab firm to ARM manufacturers, as they'll still be producing an inferior product.  However, if it does this, it would slow the growth of its own mobile product -- superior or not -- by giving ARM manufacturers access to its industry-leading process.

So this is the only case where I see Mr. Brookhaven's prediction as possibly holding true, and even if this is the situation, it's still a big IF that Intel would sacrifice growth for short-term profits.

second scenario is that x86 and ARM are roughly tied, assuming a comparable process.

In this case Intel has a lot to lose by given ARM chipmakers access to its process.  If it doesn't it has an advantage (as again the above assumption is architecture equality, not process equality).  If it does, that advantage vanishes.  So in this circumstance it's extremely unlikely that Intel would play fab for ARM chipmakers.
 
Bay Trail -- chipset
Bay Trail has impressed, thus far, hinting Intel's architecture may not be as bad as some believe it to be.

third scenario is that x86 is inferior to ARM at a comparable process, but comparable or superior to ARM when the x86 chip is built using a superior process.  From the benchmarks I've seen to date, I personally believe this is most likely.

Compared to the latest generation of 28 nm ARM chips, the 32 nm Atom CPUs proved somewhat inferior in both battery life and computational power in benchmarks from reputable sources, such as AnandTech.  Bay Trail, though, built on a 22 nm process manages to narrowly beat 28 nm ARM chips in battery life and compute performance (see, for example AnandTech's benchmarking of the Transformer Book T100 from ASUSTek Computer Inc. (TPE:2357) against the Samsung Galaxy Note 10.1 2014 Edition).

In this scenario Intel has the most to lose if it plays fab, as it's gone from a potential narrow win to a loss by handing the ARM OEMs its key advantage.  Again the benchmarks seem to suggest this is the scenario Intel is in, so it seems moderately unlikely that Mr. Brookhaven's prediction will prove true.

fourth scenario is that x86 is so drastically inferior to ARM architecturally that a process lead by Intel can't make up for it.

This is perhaps the most interesting scenario, in the sense of thinking of how Intel would react, if not overly likely.  If Intel were faced with this scenario, I believe Intel would simply bite the bullet and start making ARM chips, leveraging its process lead to become the dominant ARM chipmaker.  To make up for the revenue it lost, paying licensing fees to ARM Holdings, it could focus its efforts in the OS space (it's Tizen Linux OS project with Samsung hints at that).  Or it could look to make up for lost revenue by expanding its production of other basic process-sensitive components (e.g. DRAM).  I think this would be Intel's best and most likely option in this scenario.

But if on the odd chance this fourth scenario does occur, it would be utterly unbelievable for Intel to agree to play fab for ARM chipmakers, as it would enter one of the lowest profit, highest risk spaces, without having any real gains from it (by differentiating its higher profit offerings).

VI. Why Intel is Unlikely to Play Fab For ARM Chipmakers (Even if ARM is Better)

I understand that many believe that x86 is getting a little long in the tooth and is fundamentally less efficient in terms of power and multitasking than recent ARM architectures.  Critics point to ARM's dominance of the tablet market, which Intel has repeatedly tried to sell on the idea of x86 mobile chips, with little success.

I don't think that market dominance alone says this for sure, but I do think there's evidence for this being the case if you look at the benchmarks.  At the same time I think the advantage of ARM over x86 in the mobile/power-sensitive space is slimmer than you might think.

Intel process progress
Intel still has a major edge in process technology.

It will be very interesting to see how Intel's 22 nm smartphone chips (Tangier (platform: Merrifield)) will perform.  As mentioned, 22 nm tablet product looks very competitive; the smartphone chips arrive later but may be even more of a winner as the have a 22 nm LTE modem, giving Intel the lead in a critical component to battery life.

Looking at the four scenarios of how x86 and ARM may stack up architecturally, we see that under only one scenario is there any possibility of Intel playing fab for ARM chipmakers whose products compete with its own.  And even in that scenario it's a big "IF" as the approach has adverse impacts for Intel.

Unknowns aside, factoring in what we do know, thus far, about Intel's 32 nm and 22 nm mobile SoC performance, I think it's very unlikely that the one scenario where there is a possibility of Intel playing ARM fab is true.  Thus expect Intel to likely stick to its x86 guns.  And if on the odd chance those guns fizzle?  Expect Intel to opt to build ARM chips, a tenable if undesirable route, before it would make the suicidal move of playing fab for ARM.
Haswell shot
Why would Intel give up its best weapon -- its leading process technology?

Take Altera's relationship with Intel as more of an oddball collaboration than a sign of things to come.  If Intel plays third party fab for anyone it's mostly likely to be DRAM manufacturers or signal processing/FPGA chipmakers (like Altera) -- markets that Intel does not currently sell much direct product to.

VII.  But if You Want to Speculate... What About Intel Playing Fab for AMD's x86 PC Chips?

But if you want to speculate in a somewhat more principled sense, I think much more likely than Intel playing chipmaker for ARM optimizers like Qualcomm and NVIDIA is an even more shocking move to PC enthusiasts -- Intel offering to fabricate Advanced Micro Devices, Inc.'s (AMD) x86 consumer CPU product, or -- even more likely -- GPU product.
 
R7 and R9
Imagine Radeon built on Intel's process technology.  It's actually more likely than a tieup with ARM rivals, given that Intel doesn't make a discrete GPU -- not yet, at least.
 
Note, unlike Mr. Brookhave,n I'm not saying this will happen or Intel wants to do this.  But it is possible.

An AMD x86 tieup in the consumer CPU space would be a historically shocking partnership, but given AMD's small size and its on-the-fence position in the x86 v. ARM debate, it could have some benefits in terms of circling the wagons and scoring some profits.  AMD is admittedly open to manufacturing on other fabs other than GlobalFoundries -- which it used to own.  

While so far it's only considering fabs on the so-called "Common Platform Group" (e.g. Samsung, GlobalFoundries, etc.), I'm certain it would jump at the opportunity to stick primarily with x86 in exchange for being built by Intel if on the odd chance Intel chose to make this offer.

Again, I see this as somewhat unlikely for the same reason as the ARM fabrication offer is unlikely -- Intel would be offering to build devices that compete with its own products (something it isn't doing with the Altera deal).AMD
Playing third party fab for AMD would be risky, but arguably less so as Intel already dominates the market the third party partner's products compete in.
 
But if AMD is willing to pay enough, it seems more likely Intel might be willing to take that gamble as it it would be boosting a much smaller competitor in a market it currently dominates, and could relatively quickly pull the plug on that offer, should AMD gain too much steam.  By contrast offering help to its rivals in a market it's struggling to gain ground in -- the mobile space -- would be a decision with far more downsides.  For the same reason, I believe any possible tieup with AMD would be restricted to only consumer CPUs, not server chips where AMD is slightly more competitive in certain niches.

Returning to the ARM vs. x86 outcomes, if scenario three or four hold true, Intel could eventually offer ARM chipmakers a similar deal -- but only after it conquers a dominant position in the market they currently control.  And that will likely take a half decade or more, if it happens at all.

Sources: Altera [press release], Forbes [Insight 64 commentary]





Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Huh?
By Hammer1024 on 11/1/2013 2:08:18 PM , Rating: 3
" ...relatively unknown chipmaker Altera Corp. (ALTR). "

Err... Altera is relatively unknown? Have you lived under a rock for 20 - 30 years, maybe.




RE: Huh?
By Imaginer on 11/2/2013 2:51:28 PM , Rating: 2
Exactly, I mostly known Altera for working with FPGA chips back in college. They are most certainly not unknown.


RE: Huh?
By dew111 on 11/5/2013 4:41:09 PM , Rating: 2
Really. They are one of THE TWO FPGA makers. Anyone who knows anything about FPGAs knows about Altera. FPGAs are in all sorts of products, and increasingly important. Admittedly, compared with Intel, Altera has very little consumer visibility.


One other possibility
By michael2k on 11/1/2013 2:12:07 PM , Rating: 3
Not mentioned is the fact that Intel will span multiple process nodes in 2014: 14nm, 22nm, and 32nm.

If it reserves 14nm for itself but farms out 22nm to the highest bidders, it can play both games. Even at the end of 2014 we should see that TMSC will still have limited availability at 20nm or 24nm, giving the following possibility:
1) Best performance/battery life of Atom at 14nm
2) Good performance/battery life of ARM at 22nm at high price, good perf/battery life of Atom at competitive price at 22nm
3) Good performance/battery life of ARM at 32nm but a great price!

Apple might be willing to pay for the 22nm advantage when they release the late 2014 A8 where others won't, while others willing to ship at 32nm might still get a good price from Intel while releasing parts still competitive with the still normal 28nm parts being offered by TSMC and Samsung.




RE: One other possibility
By hartleyb on 11/4/2013 12:01:20 PM , Rating: 2
Another possibility is intel has been doing research into making a X86/ARM processor, or a processor within a processor, or different types of processors connected via a SOC that would allow android and X86 coded OS's to run hand in hand passing off processes to the required processor as needed. This would be very bennificial in the server market where there are two very distinct advantages to X86 and ARM, but if both you could have both on the same platform then lookout.


Got profitability wrong
By purerice on 11/1/2013 12:57:02 PM , Rating: 2
Fabrication is not #5 because barriers to entry for fabrication inhibit entry to market but TSMC and Intel have that down and make more profit than ARM and AMD.

Early on Samsung offered to help produce parts for other electronics manufacturers. Samsung kept the profit and manufacturing expertise to itself then later undercut the very manufacturers who had previously outsourced to Samsung. Intel could very likely be interested in the same idea. Score a big contract and get paid to teach itself new fab techniques it can later apply to its own chips. Win-win.




By lifewatcher on 11/1/2013 3:32:42 PM , Rating: 2
I see one sure thing Intel will achieve with this move - it doesn't need to invest as much to provide fab services to 3rd parties, as the said parties would need to invest themselves to cope with Intel's manufacturing wizardry. Right now, ARM-based chips are forced to invest in fabs in order to push products to the market. With that they create an entire infrastructure (inferior or not) that competes with Intel.

Much like the mining industry China has for precious metals - they produce stuff too cheaply and efficiently, for the US to develop its own precious metal mining. By doing this, they control the market and can increase the price for their services (to a point). Intel could play the same role for chip fabrication. Those who chose Intel's fabs, will have their ARM chips made cheaper than those, who chose to develop their own process and build facilities. This way Intel will choke future development of competing chip manufacturing and know-how. If Intel is the only standing major manufacturing entity, they could demand large piece of the pie (within reason, of course).




By SPM12345 on 11/2/2013 9:23:27 AM , Rating: 2
Intel's situation is somewhere between the third and fourth scenarios. However under both scenarios, it makes sense for Intel to get into ARM processor fabrication. The point is that x86 is no threat to ARM on the mobile platform, and will not be in future. ARM will be a player on desktops and servers in future, but it will not be a threat to the entrenched x86 monopoly on desktops and servers enjoyed by Windows devices given the failure of Windows RT.

From Intel's point of view, there is an entrenched, but declining market for x86 chips because of Windows, and Intel will continue to support Atom chips (which will be required to run Windows 8 tablets), but growth on desktops will come from 64 bit desktop/server class non-Windows ARM devices - Chromebooks, Android laptops, possibly Apple's desktop products as well given they are going 64 bit ARM for their future iPhones. Even Windows has been trying to transition (unsuccessfully) to ARM. Again, the Windows server market is tied to x86, but Linux and FreeBSD servers will run on ARM as well, and ARM will take a chunk out of the server market when a decent 64bit ARM server chip is available as a result.

Intel's reasoning would be this: We have been trying to compete with ARM on mobile chipsets but we can't compete in that area with x86. We have a declining legacy x86 market, but and nobody is going to touch that, so let's keep making that. However future growth on the desktop and server markets will be from ARM devices, and we should get our share of that market now, because it is definitely coming as soon as 64bit desktop/server class ARM chips are available. At the moment it is just to keep a foothold in the non-x86 desktop/server market, but who knows in the long run, we might even get into the mobile chipset market with ARM chips.




Not new
By HoosierEngineer5 on 11/2/2013 9:53:59 AM , Rating: 2
I am pretty sure Intel was a fab for either Xilinx or Altera at one point in the dark murky past. I believe they even sold the devices.




By testbug00 on 11/3/2013 10:18:24 PM , Rating: 2
They have low fab utilization and it makes them more money to fab ARM chips with all things given.

However, given how Intel has no history of working with others at all, and also tends to move off of nodes very fast, this could present problems over a longer term solution.

That being said, Intel DOES OFFER the best yields (which comes at a cost) and arguably the best process (they have different gate stuff than TSMC, which could make TSMC process better in some cases, despite being a large node)




By s12033722 on 11/5/2013 6:30:12 PM , Rating: 2
It's obvious from the comments about Altera being a little-known chipmaker and Intel fabbing "Altera's 64-bit ARM processor" that the author doesn't know much about FPGAs. In an FPGA, the configurable logic does most of the heavy lifting. Hard processors tied into the FPGA fabric are nothing new, and while they are very helpful, they are a side feature along with many other features (high-speed transceivers, DSP blocks, clock management, etc.)of the FPGA overall. This isn't a processor that will show up in phones, tablets, and similar devices. It's not an Intel competitor in any way. They target totally different markets.




Correction
By Shadowself on 11/1/2013 2:11:59 PM , Rating: 1
Apple does not control just elements 1 and 3 (Optimizations and OS) of its phones and tablets.

While Apple has licensed the ARM architectures, it has also licensed the instruction set which allows it to develop ARM compatible chips not directly based upon or derived from ARM's architectures. Thus, to some extent, Apple controls elements 1, 2, and 3 of your list when considering solely its phone and tablet offerings.




"I want people to see my movies in the best formats possible. For [Paramount] to deny people who have Blu-ray sucks!" -- Movie Director Michael Bay













botimage
Copyright 2015 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki