The ARM inherent efficiency myth

[mrmt]

Found that on seeking alpha:

A research paper written by several researchers at the University of Wisconsin, and presented at the IEEE International Symposium on High Performance Computer Architecture, came to the following conclusions with respect to the ARM/X86 myth:

Performance differences are generated by ISA-independent microarchitecture differences
The energy consumption is again ISA-independent.
ISA differences have implementation implications, but modern microarchitecture techniques render them moot; one ISA is not fundamentally more efficient.
ARM and x86 implementations are simply design points optimized for different performance levels

==========

The paper:

http://research.cs.wisc.edu/vertical/papers/2013/hpca13-isa-power-struggles.pdf

 

[Revolution 11]

At this point, there has been so much data refuting the ARM efficiency myth and so many people ignoring that data on a consistent basis, that I don't think this study will really change anything in a substantial way.

I do like having another study busting this myth, good work.

 

[TuxDave]

I saw this earlier this week. I already knew the conclusion but I was pretty impressed with the methodology.

 

[Vesku]

I always thought the Transmeta company's chips were in a way demonstrating that increases in overall computing power (density+speed) minimizes ISA impact.

 

[Fjodor2001]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

 

[blastingcap]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

Check out ARM vs Intel chip prices.

 

[Homeles]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

There are dozens of obvious reasons. You couldn't stumble upon any of them?

 

[Vesku]

1. Price
2. x86 chip development has for decades been focused on increasing performance where as ARM was being used in areas where performance was secondary to low power consumption.
3. Even as some x86 chips are being designed for low power consumption they are still expected to be able to run the current Windows OS
4. Less restrictions on how you use your volume discounted chips.

A15 closer than ever to rising up to mirror the lowest power x86 Intel/AMD chips.

 

[ChronoReverse]

The other thing is anyone can license ARM and build their own chips. With Intel you have to buy Intel's chips.

We've already reached the overlap point proving it was a myth. Some people just refuse to see it.

 

[sequoia464]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

Nice articles from Anand about this here - http://www.anandtech.com/show/6529/busting-the-x86-power-myth-indepth-clover-trail-power-analysis

and here .. http://www.anandtech.com/show/6536/arm-vs-x86-the-real-showdown/14

I had a Clover Trail tablet for a while and it gets amazing battery life.

 

[hans007]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

because they are low power and good enough. plus ARM actually licenses their core, meaning anyone can make an ARM cpu. intel's edge is at the highest end for computing speed. you jus tdont need that much anymore. if an ARM cpu uses the same wattage and is a little slower, but still fast enough but costs less and has a huge software advantage its gonna get used.

we have rockchip and chinese companies making ARM cpus etc, and since all ARM devices are ISA compatible it makes the hardware platform that much more dominant.

i would imagine the margins on ARM chips are a lot lower than intel chips given its a much more competitive space, but for the consumer thats good.

i mean if you could buy a 180hp inline 4 engine for $100, or say a $2000 hybrid power train that has a v8 and gets 500hp that has the same mpg, which would get put in most cars? the 180hp inline 4, because it costs less.

 

[Imouto]

i mean if you could buy a 180hp inline 4 engine for $100, or say a $2000 hybrid power train that has a v8 and gets 500hp that has the same mpg, which would get put in most cars? the 180hp inline 4, because it costs less.

And people doesn't need a train for their daily stuff.

 

[Idontcare]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

I see you got quite a few responses to this post already but I wanted to respond as well.

The prevalence of ARM-based ICs in mobile applications is not so much because ARM is (or was) a power-efficient ISA but more because "the price was right".

The article referenced in the OP does not account for "cost to reach performance parity".

ARM may not have any higher efficiency than x86 but it doesn't cost the same to develop an ARM based product as it does to develop an x86 based product.

For modern time ARM vs x86 there are a couple engineers on this forum who can speak to the R&D delta between ARM and x86 product development, but my own experience dates to the mid-nineties when TI (Texas Instruments) was making both.

We were producing top-speed 486 processors (100MHz, including mobile cpus for TI branded laptops) as well as making ARM-based ICs for cracking open the fledgling mobile phone market.

Ultimately we bailed on x86 and struck out to make tens of billions selling OMAP (ARM-based ICs) chips not because we couldn't make x86 (we had all the capability inhouse), rather the choice not to pursue x86 was because the development cost itself was not at parity with the development cost of an ARM-based product.

And since development cost must be amortized into the sell-price of the chip once it hits the market, it doesn't make sense to develop a low-power mobile x86 chip like Atom to compete with ARM if developing the Atom product costs 2-3x more than it takes to develop the very ARM product to which it is competing with.

Intel started the Atom project internally how many years ago? Probably 10yrs at least, and it took just as long to get it to parity with ARM products that weren't 10yrs in the making. And at what cost?

So long as creating ARM products remains cost-competitive to creating non-ARM products, the existing ARM licensees will continue to renew their licenses. What other option do they have?

Intel has a choice, as does AMD, but Qualcomm doesn't, nor does Google or Apple. So these mega-billion companies have no choice but to make ARM successful or die trying.

But if it is anything like it was 20yrs ago, the development cost for equivalent products favors the ARM-based IC for low-power small die size products and that will result in a selling-price advantage for the ARM chip once those development costs are amortized. (something the original article doesn't touch, financials)

 

[Charles Kozierok]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

Ask that again in five years.

Nobody has ever lost money betting against predictions of the demise of x86.

 

[Ferzerp]

Nobody has ever lost money betting against predictions of the demise of x86.

To be fair though, it only takes once. I'm not saying I think it will happen any time soon or that it is in anyone's best interest for it to do so. The critical mass of R&D benefits us all.

 

[mrmt]

For modern time ARM vs x86 there are a couple engineers on this forum who can speak to the R&D delta between ARM and x86 product development, but my own experience dates to the mid-nineties when TI (Texas Instruments) was making both

IDC, regarding this delta, do you have an idea of the current value? Is ARM currently much cheap to develop, or ARM was cheaper because it was simpler (smaller scope)?

 

[Wall Street]

IDC, regarding this delta, do you have an idea of the current value? Is ARM currently much cheap to develop, or ARM was cheaper because it was simpler (smaller scope)?

I think that ARM may be cheaper to develop because ARM doesn't make any chips itself so they would gladly send you some chip schematics and tell you how it all works for a price. If you make x86 processors and ask Intel what the best way to organize a chip would be, they wouldn't tell you anything. ARM views the architecture as the product they are selling while Intel views the chips as the product they are selling and the architecture as the secret sauce.

 

[Fjodor2001]

Check out ARM vs Intel chip prices.

Perhaps a valid argument for low-end phones where you're on a limited budget, so the CPU makes up a large part of the total cost. But it's not valid for high-end phones costing $700. If Intel would have really been a better option then it would be worth spending some extra $$$ on an Intel CPU. But still ARM is used in practically all those high-end phones.

Also, the only case where Intel can compete with ARM in low power use cases is when there's a race to idle. However once idle (which is where the CPU spends most of its time on a mobile device), ARM wins every time.

 

[krumme]

Perhaps a valid argument for low-end phones where you're on a limited budget, so the CPU makes up a large part of the total cost. But it's not valid for high-end phones costing $700. If Intel would have really been a better option then it would be worst spending some extra $$$ on an Intel. But still ARM is used in practically all those high-end phones.

Also, the only case where Intel can compete with ARM in low power use cases is when there's a race to idle. However once idle (which is where the CPU spends most of its time on a mobile device), ARM wins every time.

The days of uber expensive phones are fading, ask Apple shareprice and suppliers. People still want 1080p, preferable expensive popping Oled and better camera, preferably 42Mpix.

Intel is 7 years to late for the good profit. Good luck fighting 16 core little-big successor here (in 10% of the phone market), and Intel will have way to much capacity unless they change their business model. I am not saying they have a choise to do otherwise- it cant see it -but it doesnt look like good business to me.

I am tired of hearing all this bashing of ARM efficiency. Talk to me like i am the decisionmaker at Samsung (or Sony) and convince me to buy Intel cpu for my next phone. What is the arguments?

 

[Charles Kozierok]

I am tired of hearing all this bashing of ARM efficiency. Talk to me like i am the decisionmaker at Samsung (or Sony) and convince me to buy Intel cpu for my next phone. What is the arguments?

Right now, there probably isn't one. But in a few years, things may well be different.

A company like Intel is sort of like an aircraft carrier -- big and powerful and not really able to change course quickly. They had not put enough resources into developing low power chips before a few years ago. But they are now.

Their success is not guaranteed. But I wouldn't count them out, either. This is 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. Intel's challenge is to translate that into business success.

 

[Idontcare]

IDC, regarding this delta, do you have an idea of the current value? Is ARM currently much cheap to develop, or ARM was cheaper because it was simpler (smaller scope)?

I'd put the number at "roughly half" with huge confidence-limit bars on that of course.

To push an ARM-based OMAP product into mobile phones instead of pushing an x86-based "OMAP" product into mobile phones in the late 90's was about a 2:1 development cost delta, looking at being designed for the same node (our own), etc. That's why TI went the route it went back then.

Does such a delta still exist? I personally have no idea, but I know folks on this board who do have personal experience to speak on the matter, just a question of whether they feel they can (corporate policies and such).

But...the point I would make is that the delta, if any does exist, is immaterial in making the choice of "ARM vs x86" because outside of a few players like TI who hold/held licenses to both, the big players in ARM right now do not have the option of dropping it in favor of developing an x86 alternative.

For them, Qualcomm and Samsung and others, they either make ARM work or they become roadkill on Intel's bumper.

Or, to put it differently, the debate is largely academic at this juncture. Like that of "RISC vs CISC".

 

[beginner99]

If it's a myth, then how come basically everyone is using ARM CPUs in low power devices such as mobile phones?

The Razr i has longer battery life than the Razr M. The i ships with an intel, the M with an ARM. Display and the whole chassis of the phone is the same.
The Razr I also beats the M in terms of CPU performance.

 

[StrangerGuy]

Perhaps a valid argument for low-end phones where you're on a limited budget, so the CPU makes up a large part of the total cost. But it's not valid for high-end phones costing $700. If Intel would have really been a better option then it would be worth spending some extra $$$ on an Intel CPU. But still ARM is used in practically all those high-end phones.

Also, the only case where Intel can compete with ARM in low power use cases is when there's a race to idle. However once idle (which is where the CPU spends most of its time on a mobile device), ARM wins every time.

Why would you want Intel when consumers doesn't demand them AND you get to pocket more profit especially if you are the one making your own chips like Samsung?

We still haven't seen battery life tests for the supposedly ultra-efficient A7 28nm SoCs that are already shipping like the MT6589 and the upcoming big littles. That will be the real fight for Intel, not winning over some dog chip like Tegra 3 that was already terrible by ARM standards.

 

[Arachnotronic]

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.

 

[Vesku]

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.