ARM "Ecosystem" Heath

[dealcorn]

I have long struggled to understand the argument that the healthy ARM ecosystem produces fierce competition ensuring the best of the best. The argument has appeal because the word ecosystem sounds scientific. The image of life coming together scientifically to guarantee your product is the best carries a big emotional resonance. Still, I have to step back and challenge whether this makes sense.

A healthy ecosystem produces competition for scarce resources and descent with modification. Over time, the giraffe's neck gets longer so it can eat leaves from branches higher in the tree. In low power, mobile Soc world, that means performance per watt (efficiency) grows. I do not dismiss cost as a factor, but the pivotal survival metric in the niche has long been identified as performance per watt and no one proposes a credible, alternate metric.

In Anand's Silvermont Archiceture article he reported, but did not verify, Intel's claims that Baytrail efficiency is vastly superior to anything quad core ARM is expected to have at launch time. Intel's position is, in niche, any reasonable workload, any credible measurement, we dominate efficiency when compared to Intel's expectations for ARM's best quad cores at release time.

Perhaps, everything Intel says is "Liar Liar pants on fire" stuff. However, Anand presented a credible explanation why Intel's position is reasonable and Anand's annecdoctal, hearsay evidence says that some of Intel's representations are spot on. If Intel's efficiency position is later, independently verified, the characterization of what ARM does as a healthy ecosystem is nonsense.

Has ARM ever faced a competitive market? To me it looks like ARM is a single source supplier (monopolist) that uses many partners to package what is substantively the same product properly tweeked to make everone special. It is not surprising that the single most important survival metric would stagnate in such a non competitive environment.

If you love the metaphor too much to drop a loser, think of Baytrail as an invasive species that becomes dominant. I prefer characterizing Baytrail as better engineered by smart folks. You may also retain a pixie dust and unicorn metaphore but do try to preserve some correlation with the real world.

My point is real words should have real meanings. It is not helpful to shout "meh" everytime reality is harsh. Rather than capturing your sense that you are a current, with it, kinda guy, I interpret the use of the work "meh" as characterisitic of denial with a strong commitment to brain death. However, that may be a generational thing in that the site serves diverse audiences

 

[NTMBK]

ARM is far from a single source supplier. There are the standard ARM cores used by many (Cortex A9,A15) but there are also totally custom cores, like Apple's Swift, Qualcomm's Krait, and NVidia's Denver. Not to mention a lot of what makes a good SoC is not the core, but the uncore.

 

[ShintaiDK]

Its just further developments of the ARM design that ARM supplies. ANd thats essentially the whole weakness in ARM. The little ARM design company is essentially who will battle Intel. And also why 64bit for ARM is so late.

Companies like Samsung and Apple would also ditch ARM in a heartbeat if Intel delivered something better. And we are seeing Atom getting into Samsung devices now.

 

[Cerb]

Competition can breed excellence, but can also simply weed out those with less deep pockets. Lack of competition, OTOH, basically always breeds laziness, regardless of technological advancement (IE, making something better in ways that aren't the ways the customer wants).

ARM is hardly a monopolist. They're offering exactly what these other companies want: a compatible ISA, to reduce costs when using closed hardware. They also license CPU and GPU designs, but note that Apple, Samsung, Qualcomm, and Marvell, at least, typically use their own CPUs. But, by using the shared ISA, they can also be held back. For example, ARM is only just now going into supporting 64-bit, and virtualization.

With Intel, the big question is how much they are going to be selling them for. If they can't produce them cheap enough, they'll be fairly expensive, in comparison, and not offer any major threat. If they are cheap enough, they could cause quite a disruption. I wholly expect that higher-end devices will end up with Intel chips in them, either way, though.

 

[StrangerGuy]

The ARM is ecosystem is insanely competitive by itself even without Intel in the picture, just look at how the chips evolved from the ones used in the original iphone to the S600s in a 6-year span. Chip makers like TI that left ARM is no indication of how ARM is failing but on the contrary describes how cutthroat their market is.

Don't let me started on a concept called "good enough", too...

 

[NTMBK]

Its just further developments of the ARM design that ARM supplies. ANd thats essentially the whole weakness in ARM. The little ARM design company is essentially who will battle Intel. And also why 64bit for ARM is so late.

The only part of the "ARM design" which Swift and Krait use is the ISA- everything else is completely different...

 

[Imouto]

Companies like Samsung and Apple would also ditch ARM in a heartbeat if Intel delivered something better. And we are seeing Atom getting into Samsung devices now.

This again...

Samsung ain't expanding its foundries to buy chips from Intel. Samsung isn't developing the Exynos SoC to buy chips from Intel. They're buying Snapdragons and they haven't stopped development or foundry expansion. They just use/buy the most suitable solution while expanding.

Samsung is developing its own OS to get even more independence. I don't see them chaining themselves to Intel or anyone else.

 

[Arachnotronic]

This again...

Samsung ain't expanding its foundries to buy chips from Intel. Samsung isn't developing the Exynos SoC to buy chips from Intel. They're buying Snapdragons and they haven't stopped development or foundry expansion. They just use/buy the most suitable solution while expanding.

Samsung is developing its own OS to get even more independence. I don't see them chaining themselves to Intel or anyone else.

They're developing this OS in collaboration with Intel

 

[sm625]

Intel is pretty much a nonfactor because their parts are so damned expensive. There is a dearth of tablets for right around $150, and that is probably going to be a critical price point going forward. Intel isnt going to have any dog at all in that fight. Intel will have to do some serious bribing to get a vendor to use a $60 atom chip instead of a $30 qualcomm chip, even if the atom gives 40% more perf/watt, what does it matter when it is 100% more expensive?

 

[mrmt]

Intel is pretty much a nonfactor because their parts are so damned expensive. There is a dearth of tablets for right around $150, and that is probably going to be a critical price point going forward. Intel isnt going to have any dog at all in that fight. Intel will have to do some serious bribing to get a vendor to use a $60 atom chip instead of a $30 qualcomm chip, even if the atom gives 40% more perf/watt, what does it matter when it is 100% more expensive?

What makes you think Intel is selling $60 atom chips for phones and tablets, and that Silvermont will cost you that money?

 

[R0H1T]

What makes you think Intel is selling $60 atom chips for phones and tablets, and that Silvermont will cost you that money?

At similar performance levels an Intel chip is anywhere between roughly a quarter to double more expensive in price as compared to competing ARM chips which is an undeniable fact, unless you can prove otherwise can you ?

 

[dealcorn]

No one questions that ARM customers compete intensely between themselves. They do all build on a foundation that comes from the mind of ARM and that mind has not been subject to fires of competition in this space. That is why the ARM is so vulnerable if Intel actually delivers superior efficiency. Efficiency and cost largely determine survival in this space. I see a simple world. If Intel actually delivers superior efficiency and most ARM customers withdraw from this space due to inability to compete profitably, I call that death. An ecosystem where almost everyone dies can not be characterized as healthy.

However, NTMBK is spot on stressing the importance of the uncore. Lack of uncore skills have held Intel back from a broader deployment of Atom from day 1. In the Silvermont Architecture article Anand cried in his beer asking why Intel did not do more with Saltwell at 32 nm. Give me a break. Intel did not have all the uncore stuff sorted out and Intel did not want to release mockable, me too products where Intel was marginally better or worse than competitors. In almost the same breath, Anand questions whether doubling Saltwell's performance (BayTrail) is good enough to run Windows credibly. Saltwell Linux graphics support at 1080p is worthless in say, my n2800. Intel has had some failures and domination of the mobile SoC markets is critical to Intel's ability to maintain its foundry model. Intel must sell more transistors to afford increasingly more expensive fabs. Given the importance of the mobile SoC markets to Intel's survival, Paul Otellini went into drunken sailor spending mode to ensure Intel had all the pieces lined up for rapid market share dominance at reasonable profit margins. I suspect Paul succeeded, but consumers get the final word.

While Intel is on the left coast, they know where Madison Avenue is. Intel will effectively communicate to consumers why they should demand Intel inside for superior battery life. If Brand A does not have Intel inside, Brand L will. Never underestimate the power of fear and panic over possible market share loss as a motivator for device manufacturers.

Intel has a pretty good act going with China, Inc. Intel likely explains how China, Inc. can exploit Intel's reputation to expand their brands reach beyond Asia to the developed world while retaining Chinese ownership of the brand name. As long as China is already doing the assembly work, China should also own the brand names. It is only fair. Given the speed with which Android tablets stole market share from Brand A, they understand the risk.

I wonder who funded HTC's acquisition of rights to Liquidmetal. Anyone who deploys LiquidMetal technology for a phone case would be an idiot not to to spend what it costs to get a prestigous design museum to award the the title of world's best designed phone to their Liquidmetal product. I must assume Intel would like to be inside that phone. Apple bought a perpetual, exclusive worldwide license to Liquidmetal technology that locked everyone else out. Suddenly HTC has production rights and Intel and Apple are all kissy face over Haswell dual cores. HMMMMM.

While I question ARM's ability to maintain low power mobile SoC market share, ARM has a product portfolio that is broad, diverse, and largely not contested by Intel. Outside the low power, mobile Soc segment, ARM should fear China, Inc more than Intel. Further, based on their business/financial structure, as a viable entity, ARM is darn near invulnerable. ARM will not go down even if some market share is lost.

I am surprised Intel is able to get anything done at 32 nm. Intel's scattered successes in emerging market smartphones at 32 nm are incredible. How can Intel compete in this market unless it is more cost competitive at 32 nm than folks think. ARM should have dominated based on lower cost. If Intel can succeed in this segment at 32 nm, I do not see Intel's ability to compete based on cost at 22 nm as a credible issue.

 

[kimmel]

At similar performance levels an Intel chip is anywhere between roughly a quarter to double more expensive in price as compared to competing ARM chips which is an undeniable fact, unless you can prove otherwise can you ?

I don't believe you can't prove pricing that OEM/ODMs will pay or have paid for Atom chips. So stating this pricing advantage as undeniable fact without any actual facts is rather disingenuous.

 

[mrmt]

At similar performance levels an Intel chip is anywhere between roughly a quarter to double more expensive in price as compared to competing ARM chips which is an undeniable fact, unless you can prove otherwise can you ?

Undeniable fact? To assume something as undeniable you must first *establish* this something as a fact. Did you or someone here? Where's Intel OEM pricing? How much Motorola and Samsung are paying for Intel mobile chips?

 

[dealcorn]

Duplicate post.

 

[R0H1T]

I don't believe you can't prove pricing that OEM/ODMs will pay or have paid for Atom chips. So stating this pricing advantage as undeniable fact without any actual facts is rather disingenuous.

I guess that is true to an extent, certainly wrt my own OTP comments, but you can have cheap quadcore ARM(cortex A7) based phones for as low as ~125$ while the cheapest Atom based phone is roughly twice its price so taking that as a starting point I'm assuming the Atom SoC itself is fairly expensive as compared to the cheapest ARM SoC in that mid/low range performance level category. Yes I know its a stretch to claim such stuff but there is a reason why clovertrail(+) chips are found in high end phones only.

 

[R0H1T]

Undeniable fact? To assume something as undeniable you must first *establish* this something as a fact. Did you or someone here? Where's Intel OEM pricing? How much Motorola and Samsung are paying for Intel mobile chips?

That would depend on their negotiating power, as for the rest of it I've posted a reply above but I'd definitely retract my previous statement(the undeniable fact part) provided you have actual numbers to prove your claim that they don't cost twice as much even in rarest of rare cases.

 

[mrmt]

I guess that is true to an extent, certainly wrt my own OTP comments, but you can have cheap quadcore ARM(cortex A7) based phones for as low as ~125$ while the cheapest Atom based phone is roughly twice its price so taking that as a starting point I'm assuming the Atom SoC itself is fairly expensive as compared to the cheapest ARM SoC in that mid/low range performance level category. Yes I know its a stretch to claim such stuff but there is a reason why clovertrail(+) chips are found in high end phones only.

And what does A7 prices tell us about A15/Krait/A6/Exynos prices?

 

[mrmt]

That would depend on their negotiating power, as for the rest of it I've posted a reply above but I'd definitely retract my previous statement(the undeniable fact part) provided you have actual numbers to prove your claim that they don't cost twice as much even in rarest of rare cases.

Why should *I* have to prove something here? It's you who is claiming something here after all, not me.

 

[Exophase]

Intel has a pretty good act going with China, Inc. Intel likely explains how China, Inc. can exploit Intel's reputation to expand their brands reach beyond Asia to the developed world while retaining Chinese ownership of the brand name. As long as China is already doing the assembly work, China should also own the brand names. It is only fair. Given the speed with which Android tablets stole market share from Brand A, they understand the risk.

It's funny you even bring up China, since the market is now flooded with dirt cheap tablets (and PMPs, some phones even) running Chinese SoCs. Almost all of these use ARM CPUs, and this is possible because ARM licenses cheap hard-macros tweaked for Chinese fabs. Intel will have a hard time competing in a market that heavily favors these locally made SoCs.

On the other hand, Intel has pushed some very cheap Lexington based phones into Africa, probably reference phones fully made by Intel. I think they're taking much lower than typical margins on the phone itself, and are doing this in order to grow a developing market that others are ignoring and cement some major brand recognition. We'll see how much these efforts help them in the long run.

 

[dealcorn]

While perfect economic scaling (cost per transistor drops 50%) is no longer a characteristic of a process node advantage, Intel is likely darn close and reportedly doing better on this attribute than competitive fabs. If Intel can make the economics work at 32 mn, it looks easy at 22 nm.

ARM has a strong presence in China, but MIPS is closer to most favored nation status. I view both as more of a sub contractor role. I think China prefers that the worlds fastest super computer is theirs and they want the world's best IT infrastructure yesterday. The Government may feel a need to raise the population's standard of living to earn their legitimacy. Intel is profoundly helpful in helping them achieve their goals across multiple industries. If Baytrail burns some domestic fabs, so be it. Throughout it's history China has paid hugh costs in furtherance of it's perceived national interest. The Great Wall was not free but was necessary. If Intel can help them achieve important national goals across multiple industries, they are a friend.

As a side note, Anandtech really ought to be in China at every level. That's where much of the growth is and now is a good time to capture some eyeballs and commence the ad revenue flow. Translate what you got, hire some local talent and try to rule the world. If Anand understood the threat posed by China, Inc., I am not sure he would have said Intel needs a big win like Apple or Samsung. Fear is a wonderful motivator.

 

[Exophase]

MIPS has some standing with the Chinese government that is funding a MIPS CPU. By all indications it has never made much of a commercial impact and progress seems to have majorly dried up. So for all the noise made about it in the press I don't think it actually really matters that much. And that's not that surprising because it isn't a very good CPU. They're never going to get the world's fastest supercomputer this way, but that really is neither here nor there when talking about mobile or really anything of commercial significance.

There's a MIPS SoC vendor, Ingenic, but just that one, and they're falling way behind the Chinese ARM SoC makers. I doubt they'll be a major player going forward.

And I'm not sure you realize this but the Chinese tech market is more or less private, not government ran. No one's going to burn fabs to make way for Bay Trail. Chinese SoCs are not chasing best perf (or best perf/W for that matter), simply manufacturing prices that Intel will not be able to compete with even if they took essentially no margins on entire phones.

 

[dealcorn]

And you are aware that these fabs are also able to make SoC's for markets not contested by Intel. Is Baytrail succeeds it will kill some Chinese ARM fab jobs. In the past this might be cause to shut down Intel's access to the Chinese market. China frequently intervenes in it's marketplace to create the order it desires. It is about control, not ownership. My position is that Intel has some reasonably good, fundamental relationships that make it clear that it is far more a friend than an enemy to be feared. If Intel acts in furtherance of national interests, they will not arbitrarily hassle Intel.

If Intel can help Chinese brands gain developed world market share, that is more important than some sub contractor jobs. The profit margins are better too. The Mike Bell guy appears to have some sense of developed market preferences and he may be helpful with some aspects of China's design.

 

[Exophase]

I don't know why you think Bay Trail is so appealing to the Chinese market. I don't know where you think these relationships with the industry are. Connections with the government, if really there, has less impact than you think. It won't get Chinese OEMs to spend substantially more money on tablets and phones unless the government is heavily subsidizing them and regardless of how friendly you think they are why on earth would that happen?

Chinese SoC makers aren't sub-contractors, they're IP licensees.. not sure why you keep calling them that.. They're also doing okay gaining more and more foreign market share without Intel.

 

[mrmt]

I don't know why you think Bay Trail is so appealing to the Chinese market. I don't know where you think these relationships with the industry are.

There is no bleeding edge chinese ARM designer or manufacturer, they all use vanilla or semi-custom ARM SoCs manufactured in n-1, n-2 or even n-3 node. If they want to bring bleeding edge, they have to tap someone outside China, and if Intel comes with a good business proposition, why not?