The ARM inherent efficiency myth

[Puppies04]

The days of uber expensive phones are fading

I would love to see your proof for this, if you are just going to link to apples share price please bear 2 things in mind first.

1. Apples share price last year was an absolute joke, it was a gold rush and the only thing that has happened is that the bubble burst and the share price came back to a realistic level (until the next time some billionaires decide they want to have fun with the stock market)

2. The current share price is roughly 3 x what it was before the first Iphone was released.

As far as I am concerned the high end phone market is in the middle of a massive performance war, with faster more power efficient ARM CPUs, better screens, higher resolution cameras and better battery technology all being actively developed. The chance of this happening while the "top tier" handset price reduces by some massive % is basically zero.

 

[jaqie]

None of this makes any sense at all to me and never has... the inefficiencies are in x86 not arm, and they have been pointed out in this thread several times already. They would never be used in high power cellphones if they were that inefficient nor would x86 be having such a hard time getting into that, it would be a shoe-in for such and would have gotten itself into the market by it's technical virtues.

And this is from a stout and stalwart x86 nut er...enthusiast fangirl.

 

[LogOver]

I don't think the price issue is a myth. Intel wasn't really looking to sell chips for under $10. They sold off XScale.

I don't think Qualcomm and Nvidia sell chips for under $10.

 

[CHADBOGA]

None of this makes any sense at all to me and never has... the inefficiencies are in x86 not arm, and they have been pointed out in this thread several times already. They would never be used in high power cellphones if they were that inefficient nor would x86 be having such a hard time getting into that, it would be a shoe-in for such and would have gotten itself into the market by it's technical virtues.

And this is from a stout and stalwart x86 nut er...enthusiast fangirl.

Sounds like you either didn't read or didn't understand the paper presented in the opening post.

 

[jaqie]

Sounds like you either didn't read or didn't understand the paper presented in the opening post.

It sounds to me like you don't understand my post, or didn't read it.
Careful, don't wanna cross the line into trolling here.

See, if you were interested in countering my position, you would have quoted Idontcare's post as we have the exact same position on this issue, but he has posted far more information as he has the time and energy to do so.
Instead you quote mine and give the spurious/troll-like argument that I must not have read or understood the original article, instead of countering more specific arguments of someone else in the thread who has given more specific reasons they have the same position I do.

 

[Idontcare]

So, we go from an X86 efficiency myth to an X86 price myth? Typical

The reason ARM got there first was that Intel wasn't targeting the same performance/power levels as the ARM vendors were. Plain and simple.

The paper focuses on technical efficiency, not financial efficiency. And there is a financial difference between ARM and x86. It isn't a myth.

I don't think Qualcomm and Nvidia sell chips for under $10.

Sure the ASP for their absolute top-end SKU is going to be much higher, probably in the $60-$80 range, but the bulk of their chip sells for mobile phones are going to be sub-$20.

Consider this - in spring 2007 at 65nm when TI held 60% of the marketshare for mobile chips (down from 80% in earlier years) and we were selling a product called "UPP8M" which was a universal phone processor (UPP) with 8M of on-die cache, 12MP cameraphone capability, etc etc.

This advanced (even by today's standards) state of the art smartphone chip had a target selling price of somewhere around $40 IIRC, and that was our top-of-the-line SKU.

At the same time we had a SKU targeting the low-cost markets (free mobile phones in Africa and India) which was appropriately dubbed UPPLoCosto (not making that up :\) and it was designed with a target sale price of ~$2-$3 IIRC. And that was 2007, over 6yrs ago.

Despite those prices, TI's marketshare did not fair well (now is <5%).

 

[Charles Kozierok]

Jaqie, I'm having a hard time even understanding what your position is on this. You say there are "inefficiencies" in x86, but then say x86 wouldn't be used in cell phones if it were too inefficient. Missing something...

 

[jaqie]

I also said I share the same position as Idontcare, and that I don't have the energy or time (specifically energy, though I do now, health issues) to post in detail as Idontcare does... I thought I made it quite clear in my edited post above, I'm sorry that wasn't the case.

I don't think I said it isn't used in cellphones, but I did try to say that cellphones would be x86 instead of arm in great majority if they were the most efficient for the task.

Maybe I should put in my sig that I have advanced MSA (A.K.A. shy-drager) and that sometimes I cannot fully explain myself, sometimes for lack of energy sometimes for pain sometimes for just being so out of it because of my disorder or meds for pain that I can't make sense... The last man standing quote which is there now I admit is a bit nebulous but was an attempt at hinting at this without seeming like a pity-party.

 

[mrmt]

I don't think I said it isn't used in cellphones, but I did try to say that cellphones would be x86 instead of arm in great majority if they were the most efficient for the task

As explained before, x86 wasn't used in phones because:

1) It was cheaper to develop ARM chips

2) Not everyone in the party had a x86 license

It wasn't because ARM is more efficient than x86. Those efficiencies you see are just optimizations for a given performance levels. Put the two at the same performance levels and there won't be any edge from the ISA itself.

 

[jaqie]

As explained before, x86 wasn't used in phones because:

1) It was cheaper to develop ARM chips

2) Not everyone in the party had a x86 license

It wasn't because ARM is more efficient than x86. Those efficiencies you see are just optimizations for a given performance levels. Put the two at the same performance levels and there won't be any edge from the ISA itself.

and yet again someone takes me out of context.

Oh well, I'll answer you anyway.
Arm is more financially efficient than x86, and you are listing the very reasons it was right there. I said nothing about only or specifically ISA or isa compliant devices, I am talking about arm versus x86 efficiency for mobile phone chip use.

As an aside, it is kinda hard to get intel to either develop for you or allow you to develop an SOC solution so you are stuck making a more complex board with more duplicated parts or circuits (or both) and more power use and a larger more complex board with more support materials.....and on and on. It is also as said before by Idontcare far more dev time to write x86 code for this application (perhaps any application) than it is for ARM, especially if you have to write in tie-in code for a second chip which does system tasks that the SOC could have done without special interfacing and on and on et cetra ad nauseum.

 

[mrmt]

Oh well, I'll answer you anyway.
Arm is more financially efficient than x86, and you are listing the very reasons it was right there. I said nothing about only or specifically ISA or isa compliant devices, I am talking about arm versus x86 efficiency for mobile phone chip use.

Financial efficiency =! Performance efficiency. The myth busted in the article is about performance efficiency, not financial efficiency. And as IDC pointed out, financial efficiency is a moot point. Not only ARM chips grew in complexity, ARM manufacturers can't go for x86,

As an aside, it is kinda hard to get intel to either develop for you or allow you to develop an SOC solution so you are stuck making a more complex board with more duplicated parts or circuits (or both) and more power use and a larger more complex board with more support materials

I think you are mixing the things here.

What you are describing is Intel selling too much performance for you, and that you, for cost reasons, need a smaller, lower performance processor for your application. In this case, Intel isn't interested in you, because Intel is a company built around bleeding technology, and you don't need bleeding edge. This has nothing to do with financial efficiency, it's just that what you need isn't what's on the shelf to buy.

We might have an answer for x86 financial efficiency in a few years, as there is a x86 manufacturer making inroads in the low cost embedded market. We'll see how financially efficient x86 can be for the low cost market if AMD survives long enough with its embedded business.

 

[jaqie]

Financial efficiency =! Performance efficiency. The myth busted in the article is about performance efficiency, not financial efficiency.

The OP (especially the title!) mentions nothing about this thread being about solely performance efficiency, but even if it did the performance efficiency of x86 is extremely lackluster for this market because you aren't getting x86 SOC, as I pointed out, that makes the system performance in this application very bad once you add in all of the space, board layout, support components, power draw, extra programming.... et cetra needed for such.

And as IDC pointed out, financial efficiency is a moot point. Not only ARM chips grew in complexity, ARM manufacturers can't go for x86,

Far from moot unless the market is willing to bear the extra costs and major loss of efficiency of having to run a chipset solution instead of a SOC solution. Judging from market trends, it isn't. If it was, the industry would most definitely develop into and optimize x86 based solutions! It really is that simple.

I think you are mixing the things here.

What you are describing is Intel selling too much performance for you, and that you, for cost reasons, need a smaller, lower performance processor for your application. In this case, Intel isn't interested in you, because Intel is a company built around bleeding technology, and you don't need bleeding edge. This has nothing to do with financial efficiency, it's just that what you need isn't what's on the shelf to buy.

This has nothing to do with me. If it was, I would be buying x86 stuff whenever I could. Oh, wait, I do. And when there is nothing x86 in the market with the features I want, I must go with what's available. Woe is me, it's almost entirely ARM. At least I can and did get a blackberry and stay out of the whole ios versus android mess, at least for a while. I would so love me a tablet that is x86, but I can and do have one! It's called a tablet PC! Oh, happy day. Too bad most people don't see it my way and won't sacrifice... what?.... FINANCES for the $1000+ tablet PC instead of a $50+ (and up) arm tablet.

We might have an answer for x86 financial efficiency in a few years, as there is a x86 manufacturer making inroads in the low cost embedded market. We'll see how financially efficient x86 can be for the low cost market if AMD survives long enough with its embedded business.

I hope so, I really do. But until then, arm is just far, far more efficient, and thus has hard-won it's current share of this market.

 

[Charles Kozierok]

This has nothing to do with me. If it was, I would be buying x86 stuff whenever I could. Oh, wait, I do. And when there is nothing x86 in the market with the features I want, I must go with what's available. Woe is me, it's almost entirely ARM.

I think you guys are talking past each other here.

Nobody is denying that right now, ARM chips are more energy-efficient. What we are saying is that this is because of design and business decisions, not anything particular to the ARM ISA versus the x86 ISA.

As Intel makes efficiency more of a priority, that gap will narrow and possibly disappear.

It's sort of like saying Porsche isn't capable of making cars that are fuel-efficient, because none of their cars get great mileage. But I have every confidence that they could design an incredibly fuel-efficient car if that was their priority. It isn't.

Intel's priority wasn't energy efficiency until a few years ago. It takes time to transition a process from one emphasis to another. Whether Intel will succeed or not remains to be seen; the most anyone is saying is that the x86 ISA itself doesn't preclude this from happening.

 

[jaqie]

. . . . the most anyone is saying is that the x86 ISA itself doesn't preclude this from happening.

This I will most heartily agree with, and it frustrates me to no end that people seem to think that means arm is superior in this regard.

The original post itself especially it's title... along with several posters in the thread seem to be saying that arm systems in their entirety are less efficient than x86 systems are for this taskset and that the quoted PDF is proof of it {full stop}.

 

[Idontcare]

I also said I share the same position as Idontcare, and that I don't have the energy or time (specifically energy, though I do now, health issues) to post in detail as Idontcare does... I thought I made it quite clear in my edited post above, I'm sorry that wasn't the case.

I don't think I said it isn't used in cellphones, but I did try to say that cellphones would be x86 instead of arm in great majority if they were the most efficient for the task.

Maybe I should put in my sig that I have advanced MSA (A.K.A. shy-drager) and that sometimes I cannot fully explain myself, sometimes for lack of energy sometimes for pain sometimes for just being so out of it because of my disorder or meds for pain that I can't make sense... The last man standing quote which is there now I admit is a bit nebulous but was an attempt at hinting at this without seeming like a pity-party.

For what its worth jaqie I'd say don't take the feedback personal in that particular way, people are just asking you questions about your position not because they want to take you to task but rather they are asking you questions about your position because they don't fully understand it (as presented) and they merely want to understand it (and you) better.

Sorry to hear about the medical situation

 

[Idontcare]

This I will most heartily agree with, and it frustrates me to no end that people seem to think that means arm is superior in this regard.

The original post itself especially it's title... along with several posters in the thread seem to be saying that arm systems in their entirety are less efficient than x86 systems are for this taskset and that the quoted PDF is proof of it {full stop}.

I interpreted it in exactly the opposite manner, that the thread is about dispelling the notion that "x86 systems in their entirety are less efficient than ARM systems are for this taskset" and that the quoted PDF is proof of it {full stop}.

But I readily acknowledge I am predisposed to interpret the article in that manner because the general mantra (myth as it were, now proven by the article) in these forums to date has been that x86 suffers an intrinsic technical deficiency (an unavoidable overhead) from "the decoder stage" that would forever prevent x86 based processors from being competitive with ARM based processors in the sub-1W space.

It is with this backstory that I assumed the article was being presented. From a "myth busted" perspective on x86's supposed intrinsic inefficiency, not from an "ARM is incapable and doomed!" perspective which is where you seem to be communicating that you think the dialogue is going (and maybe I am wrong there).

 

[ams23]

The myth busted in the article is about performance efficiency, not financial efficiency

The term "performance efficiency" is pretty vague, as there are numerous metrics for architectural efficiency. For instance, modern day ARM CPU's tend to have much better performance per mm^2 of die size area vs. modern day Atom CPU's. The modern day [in order] Atom CPU's compare more favorably in performance per watt vs. modern day [out of order] ARM CPU's, but we have yet to see a comparison to the new and much more power efficient [in order] Cortex A7 CPU. The conclusions of the article are pretty obvious that low level architectural differences are more important than ISA differences when comparing architectural efficiency of ARM vs. x86, but that doesn't mean that an ARM RISC architecture has no inherent advantages over an x86 CISC architecture in terms of reduced die size area and ease of implementation.

 

[Charles Kozierok]

It's also worth pointing out that Intel not only doesn't have to beat ARM in efficiency (in terms of battery life, etc.), they don't have to even achieve parity. A lot of people are willing to give up something in that area for x86 compatibility. It really is a big deal, and it's a strong factor in Intel's favor.

 

[ams23]

It's also worth pointing out that Intel not only doesn't have to beat ARM in efficiency (in terms of battery life, etc.), they don't have to even achieve parity. A lot of people are willing to give up something in that area for x86 compatibility. It really is a big deal, and it's a strong factor in Intel's favor.

Most consumers using handheld devices such as smartphones and tablets don't care at all about x86 compatibility. And even if one did have an x86 compatible smartphone or tablet, some legacy programs may run slow or cause significant drain on battery life.

 

[Charles Kozierok]

Most consumers using handheld devices such as smartphones and tablets don't care at all about x86 compatibility.

I'll buy that for smartphones. But tablets? No way.

 

[ams23]

This the first time I can think of where we have the potential for one semiconductor manufacturer to have a fairly significant process advantage over the others.

Actually Intel mobile CPU's have not yet had a fabrication process advantage over ARM mobile CPU's. That may change at year end or next year, but the expectation is that Intel's process technology advantage in general will diminish as companies such as Samsung, Apple, Qualcomm, etc. start to invest heavily (with NVIDIA, AMD, and others) in newer and more advanced process nodes.

 

[ams23]

I'll buy that for smartphones. But tablets? No way.

The success of iPad, Kindle Fire, Nexus 7, and Galaxy tablets compared to all Windows 8 tablets is proof that most consumers do not care about x86 compatibility in a tablet.

 

[krumme]

The success of iPad, Kindle Fire, Nexus 7, and Galaxy tablets compared to all Windows 8 tablets is proof that most consumers do not care about x86 compatibility in a tablet.

I am personal in for a Atom 14nm tablet and phone. I want my excel budget scenarios by hand, and loads of office docs, i feel more secure then

But its going to be a very small niche without any economic importance. The market have chosen. Intel was to late, period.

The situation here compared to Intel entering the server market, is that the server market was populated by expensive CPU. At the time 14nm Atom comes to mobile phones, all low and midrange is already populated by the extremely efficient non OoO quad core A7 with marginal production cost in the sub 1 usd for the core part. Its like dual core A5 in todays 40nm market. The door is closed so simple is that. Add. Samsung and others have minus strategic interest in funding Intel.

Next Intel needs some serious efficient GFX to get Atom of the ground here. Perhaps they could licence it from AMD

We tend to favor CPU performance in these forums. But for phones, the game is different, and it changed years ago, with Apple as market leader.

 

[ams23]

I am personal in for a Atom 14nm tablet and phone. I want my excel budget scenarios by hand and loads of office docs, i feel more secure then

Yes, but even in this useage scenario using MS Office as an example, x86 compatibility is not required, because MS Excel/Word/etc. is available for Windows on ARM . Microsoft's problem is that they were late catering to the ARM ecosystem (as you noted), they did not use bleeding edge ARM hardware, and they created a lot of confusion in trying to cater to both ARM (Tegra) and x86 (Clovertrail) ecosystems at similar price points for end user products. Microsoft would have been better off by renaming the OS's to Windows Consumer (designed for ARM processors, which Intel can emulate if need be to get into the lowest cost systems) and Windows Professional (designed for x86 processors and higher cost systems).

 

[Charles Kozierok]

Actually Intel mobile CPU's have not yet had a fabrication process advantage over ARM mobile CPU's.

I said "potential for"... I'm talking a few years down the road.

The success of iPad, Kindle Fire, Nexus 7, and Galaxy tablets compared to all Windows 8 tablets is proof that most consumers do not care about x86 compatibility in a tablet.

No, it actually doesn't. I only proves that, right now, the advantages of x86 are viewed as less favorable than the advantages that other products offer.

If the only difference was x86 or not, I guarantee you a large percentage of customers would go with x86. It's been that way for decades.