The ARM v.s. Intel Thing - Let's Discuss It

[Arachnotronic]

Exophase

Great insight. My belief is that Intel needs to have, by far, the most compelling hardware on a perf/watt and perf/$ basis. This will lead to more widespread adoption of the platform, and then the software optimizations will fall into place (with some "big check" writing going on, too).

It is an uphill battle for Intel, for sure, but if they are serious about winning here, they have a pretty good shot at doing it. If they are at the top of their game, I see good things. If they slack off even a tiny bit, it'll be game over in the phone space.

Luckily, in tablets Windows 8 reverses the tables and gives them the advantage over Windows RT. I don't know who in the ARM ecosystem will do what Intel is doing in the Android one to get things to work properly.

 

[Fox5]

Well, Atom executing ARM code at 40% of the same code in x86 is still impressive, especially if it does it with 100% accuracy. I don't think there's any open source emulators that achieve comparable results. QEMU is one of the best and non-native code execution is at best like 10% of native performance.

Thankfully, all my killer apps on Android are the native Google apps. If Intel can get a high end design win, I'd be ok picking up an Atom based phone, the processor is pretty negligible compared to phone design, battery life, etc. I'd probably care more about faster flash storage or more ram than the other specs.

 

[Exophase]

Exophase

Great insight. My belief is that Intel needs to have, by far, the most compelling hardware on a perf/watt and perf/$ basis. This will lead to more widespread adoption of the platform, and then the software optimizations will fall into place (with some "big check" writing going on, too).

It is an uphill battle for Intel, for sure, but if they are serious about winning here, they have a pretty good shot at doing it. If they are at the top of their game, I see good things. If they slack off even a tiny bit, it'll be game over in the phone space.

Luckily, in tablets Windows 8 reverses the tables and gives them the advantage over Windows RT. I don't know who in the ARM ecosystem will do what Intel is doing in the Android one to get things to work properly.

There's a lot of metrics. Not just perf/W, perf, and perf/mm^2 for the CPU, but lots of other integration factors and political ones. And when you get down to it, I don't think the perf/W curve of the CPU matters in the market as much as we think it does. How many reviews are even testing battery life while using things that stress the phone? What seems to matter the most is dynamic power management of the SoC and load balancing plus efficiency of the software stack. That's why iOS phones and tablets did more than fine when they still had consistently slower CPUs than the competition. Don't get me wrong, Intel is doing great here; actually I think that the combination of low power states, good turbo, and good integration of peripherals are the real win they've pulled off. But I also don't think they're going to grow a massive lead in these areas, and they're not going to capture their own new OS market like Apple.

And I just don't see Intel dominating the mobile space until they get Apple running them and Samsung makes them their flagship, etc. And I don't see these companies abandoning their massive vertical integration investments to buy Intel chips unless they fall very, very behind. Even further behind than Apple's PPC cores fell vs Intel's x86, because they were the minority player to start with. I don't see this happening any time soon. I also don't see the cheap low end Chinese SoCs giving way to Intel, I don't see how they can possibly compete on price there, particularly not for that market domestically.

What I don't get is why Intel has to dominate or "win" to be successful. There's not enough room in the market for everyone, true, and we're already seeing major players give way, but there's room for more than just one. I mean, there has been so far.

I do agree with you that right now x86 Windows 8 tablets are a more compelling buy than ARM Windows RT tablets. Microsoft's push here is a strong ally for Intel. While they looked like they were boosting the ARM ecosystem this poor contrast almost feels like they're blasting it instead. Nonetheless, the onus is still on MS to actually take enough of this market for it to matter. If someone's going to fix it it may as well be MS themselves because they're the ones staking the product on ARM; Google never staked Android on x86.

Well, Atom executing ARM code at 40% of the same code in x86 is still impressive, especially if it does it with 100% accuracy. I don't think there's any open source emulators that achieve comparable results. QEMU is one of the best and non-native code execution is at best like 10% of native performance.

Thankfully, all my killer apps on Android are the native Google apps. If Intel can get a high end design win, I'd be ok picking up an Atom based phone, the processor is pretty negligible compared to phone design, battery life, etc. I'd probably care more about faster flash storage or more ram than the other specs.

It's not 100% compatible right now, much less 100% accurate. I don't know how much the current incompatibilities are down to bugs they can fix or unsafe assumptions they're making to optimize it.

But your comparison isn't fair. Intel's emulator is a "user mode emulator." What you're comparing it to is a "system mode emulator."

Basically, a user mode emulator has a lot lower overhead because they don't have to emulate anything but the CPU and they rely on the OS to provide a safe abstraction for memory accesses. So you can access pointers directly instead of having to check and convert them. And you don't have to emulate any privileged CPU stuff; no supervisor mode, no interrupts or anything like that. And syscalls are handled natively by the OS.

In this case things the may be even lower because the interface NDK code has with the rest of the world is even more limited.

qemu has both system mode and user mode style emulation. The 10% figure you're referring to is probably while running the former. If you booted an emulated system it was definitely the former. You should check out the user mode emulation some time. There have been various papers of user mode emulators for things like PPC to x86 that get much better than 40% native performance. Like Intel's numbers this is for running stuff like SPEC2k. I don't know how much either extends to the real world.

 

[Nemesis 1]

Eye opener? LOL

Did you not see the 40nm part being compared to a 32nm chip? One approaching its end of life in favour of 28nm Tegra 4? How about the fact Atom has barely been included in a single product its THAT good. What about the companion core which is turned off because its note supported by windows which normally gives Tegra its edge in power saving? Oh thats right we are comparing windows RT. A product that basically is a flop and stands little to no chance of getting anywhere just like WinPhone 8.

Samsung is testing 14nm Process nodes and im betting that they will want to catch intel up on that front eventually.

Intel claimed they were going to be a top GPU company and they failed. They also claimed they were going to be a top SSD company and they also failed at that too.

In Fact the only thing they have ever managed to beat is AMD. Now they have to face ARM which is a company with no physical production who have partners such as Samsung, Apple, Qualcomm, Nvidia, & many more. All have significant resources and cash on hand. ARM offer the chance to make bespoke SOC's What can Intel do? They made a special ULV chip for Apple once. Thats it.

Oh and Intel wont like working on the lower margins either and that will limit them always

Samsung is at the point now with 14nm intel was at over a year ago . Also mixing 20nm parts with 14nm parts is a trick to give the impression of catching up . intel will likely be releasing 14nm parts befor or around the same time as samsung releases 20nm parts. Note intel went to 14nm on the next process not 16nm. so who is doing what to whom . When intel announced dual gates IBM and AMD said we have that to . ya 3 years latter.

intel in becoming a gpu leader . they already are . No Intels gpus will not perform as good as NV . But IGP I think intel will be way ahead after haswell. Intel Nights corner is enough to really hurt NV in the high margine arena. As IGPs become more powerful fewer and fewer Add in cards will be used forcing highher pricies in that market. As more and more people will say good enough as AMD fanbois do now with trinity. If Haswell is better than trinity which it will be . its game over they lose their base debating points . Intel adding hardware to the driver stack was really a smart move and helps with the driver stack.

you may want others to catch up but it really isn.t happening as is clearly seen by intel going a full node + @14nm . As far as SSD intel is the leader . I don't buy the fastest SSD I buy intel SSD because they are flat out the best. Who was it that brought us trim who was its that brought us trim for Raid 0 Intel. So who leads the biggest seller or the guys that make it all work . INTEL

Your intel hatered is clear as it clouds the reality of the market . Intel Vs ARM is a non event . Its over at 22nm Atom . Arm will be down to childs toys. These dang phones are just a couple of steps away from full compute on the wrist Dick tracy style. Samsung is starting to beat Apple in the phone market . I fully expect to see Apple go strictly intel at @22nm. along with google on the top models arm will be for the obama phones right where they belong

 

[Fx1]

Samsung is at the point now with 14nm intel was at over a year ago . Also mixing 20nm parts with 14nm parts is a trick to give the impression of catching up . intel will likely be releasing 14nm parts befor or around the same time as samsung releases 20nm parts. Note intel went to 14nm on the next process not 16nm. so who is doing what to whom . When intel announced dual gates IBM and AMD said we have that to . ya 3 years latter.

intel in becoming a gpu leader . they already are . No Intels gpus will not perform as good as NV . But IGP I think intel will be way ahead after haswell. Intel Nights corner is enough to really hurt NV in the high margine arena. As IGPs become more powerful fewer and fewer Add in cards will be used forcing highher pricies in that market. As more and more people will say good enough as AMD fanbois do now with trinity. If Haswell is better than trinity which it will be . its game over they lose their base debating points . Intel adding hardware to the driver stack was really a smart move and helps with the driver stack.

you may want others to catch up but it really isn.t happening as is clearly seen by intel going a full node + @14nm . As far as SSD intel is the leader . I don't buy the fastest SSD I buy intel SSD because they are flat out the best. Who was it that brought us trim who was its that brought us trim for Raid 0 Intel. So who leads the biggest seller or the guys that make it all work . INTEL

Your intel hatered is clear as it clouds the reality of the market . Intel Vs ARM is a non event . Its over at 22nm Atom . Arm will be down to childs toys. These dang phones are just a couple of steps away from full compute on the wrist Dick tracy style. Samsung is starting to beat Apple in the phone market . I fully expect to see Apple go strictly intel at @22nm. along with google on the top models arm will be for the obama phones right where they belong

i love how you cram your own version of facts alone with your fortune telling.

Apple is just as likely to completely ditch intel as they are to go with them. Last time i checked though iOS didnt run on x86. Also Apples ARM SOCs are pretty fast currently.

Intel has about 12-18 months ahead on process node and intel uses its best nodes for its desktop and laptop parts. Thats why Atom is at 32nm.

Intel will offer a all in one SOC maybe but they will have to sell it at $20 or it wont sell because thats the market price. Thats a long way from $200 retail prices of its desktop chips.

Right now the only partner intel has in this space is Motorola in 1 phone. Thats a long way to go.

Also i dont hate intel, i just dont like some of the stuff they do. Such as cheap out on the TIM on their $300 CPU's

 

[TuxDave]

Also i dont hate intel, i just dont like some of the stuff they do. Such as cheap out on the TIM on their $300 CPU's

Read this post (bottom) and let us know if you feel the same.

http://forums.anandtech.com/showpost.php?p=34053183&postcount=570

 

[Nemesis 1]

i love how you cram your own version of facts alone with your fortune telling.

Apple is just as likely to completely ditch intel as they are to go with them. Last time i checked though iOS didnt run on x86. Also Apples ARM SOCs are pretty fast currently.

Intel has about 12-18 months ahead on process node and intel uses its best nodes for its desktop and laptop parts. Thats why Atom is at 32nm.

Intel will offer a all in one SOC maybe but they will have to sell it at $20 or it wont sell because thats the market price. Thats a long way from $200 retail prices of its desktop chips.

Right now the only partner intel has in this space is Motorola in 1 phone. Thats a long way to go.

Also i dont hate intel, i just dont like some of the stuff they do. Such as cheap out on the TIM on their $300 CPU's

I don't care what apple does I sold my shares I have no stake in that company. If Apple dumps intel . That I wouldn't mind seeing . I would love to see apple at $5 dollars a share again . As software catches up with hardware the need for power and efficiency grows . Right now were at good enough for now but that will change sooner than later . Besides my sooth saying since joining this forum in 2001 has been about 95% correct. That in all subjects . Don't bother looking under my present handle as I was banned for being a soothsayer and I was unjustly banded as my words turned out to be true facts . The first ban . I kinda agree with but it was 2001 and I did make a good point for that time but yes it was wrong . Everthing else was all AMD NV fanbois raising hell because they didn't like what I had to say. To bad I was correct for the most part. My only regret is Duvie I really like the guy but he will only post here when I am banned for that I am truely sorry . The guy is ruff around the edges but was a good forum member and contributed much. I missed the part as to why Atom is at 32. Well I have to admitt what intel did with atom was to honor a 5 year commitment to its present form . Why I don't know Robert doesn't know either. At 14 nm your reasoning is blown out of the water as Atom will be released at the same time as big chip . According to intel anyway .

 

[sontin]

Samsung is at the point now with 14nm intel was at over a year ago . Also mixing 20nm parts with 14nm parts is a trick to give the impression of catching up . intel will likely be releasing 14nm parts befor or around the same time as samsung releases 20nm parts. Note intel went to 14nm on the next process not 16nm. so who is doing what to whom . When intel announced dual gates IBM and AMD said we have that to . ya 3 years latter.

And yet Intel released their latest SoC on the same process node at the same time as Samsung and Qualcomm. They don't care that Intel is ahead because Intel will never use their latest process technology from the start for the $2x market of SoCs.

intel in becoming a gpu leader . they already are . No Intels gpus will not perform as good as NV . But IGP I think intel will be way ahead after haswell. Intel Nights corner is enough to really hurt NV in the high margine arena. As IGPs become more powerful fewer and fewer Add in cards will be used forcing highher pricies in that market. As more and more people will say good enough as AMD fanbois do now with trinity. If Haswell is better than trinity which it will be . its game over they lose their base debating points . Intel adding hardware to the driver stack was really a smart move and helps with the driver stack.

And discrete gpus are not going faster with the next process shrink?
nVidia's slowest 28nm mobile card is 30-50% faster than HD4000. Haswell will only catch up with the GT640m...

Your intel hatered is clear as it clouds the reality of the market . Intel Vs ARM is a non event . Its over at 22nm Atom . Arm will be down to childs toys. These dang phones are just a couple of steps away from full compute on the wrist Dick tracy style. Samsung is starting to beat Apple in the phone market . I fully expect to see Apple go strictly intel at @22nm. along with google on the top models arm will be for the obama phones right where they belong

Wow - and yet there are more Tegra 3 phones than phones with Midfield. There are more Windows RT tablets with Tegra 3 than Windows 8 with Clover Tail. And we talking here about 40nm vs. 32nm.
Nothing will change. Intel can't compete on the same process node. They need the next one to be competitive. And that means that they will never get more OEMs than nVidia. nVidia is selling Tegra 3 for <=$25. Tegra 4 will be $30 - $35.

Intel will never get a huge chunk of the market if they not cutting the price of ATOM.

 

[IntelUser2000]

Intel will never get a huge chunk of the market if they not cutting the price of ATOM.

Your argument about price is incorrect though. The WinRT Tablets and the Windows 8 Atom Tablets cost the same.

 

[jpiniero]

Your argument about price is incorrect though. The WinRT Tablets and the Windows 8 Atom Tablets cost the same.

All that says is that the Surface RT is a bad value. Which it is.

 

[sontin]

Your argument about price is incorrect though. The WinRT Tablets and the Windows 8 Atom Tablets cost the same.

With tablets over $500 price of the SoC makes only a small part of the overall costs. But that is the high end and for that market you need the fastest SoC and the best package.

 

[IntelUser2000]

On that point, I believe Atom and Cortex-A9 have similar perf/MHz with A9 being slightly higher, but even Medfield can turbo well past the clocks we see in A9s out (and has quite good power consumption at these high clocks, props to Intel for this). A lot of people will disagree. Some are even saying Krait and Atom have clock for clock parity.

Exo,

I can believe the bolded sentence. Krait isn't that much faster than Atom, except in FP. General performance doesn't seem to be hugely than the regular A9 cores.

According to ARM though, the 1.4GHz A9 gets SpecINT2K score of ~455. The original Silverthorne based core at 1.6GHz gets 653. 1.8GHz Pineview D525 gets 725.

Acer W510 benchmark shows an advantage of 45-65% in TouchXPRT benchmark. That corresponds to the approximate difference in performance for SpecInt2K.

That difference reduces to 20-60% on the WebXPRT benchmark, which can be explained by having the application multi-threaded to some degree at least.

Atom's Hyperthreading brings ~35% in SpecIntRate. That with 60% single threaded advantage translates into ~2.2x gain, with probably quad core in Tegra bringing not double, but 1.7-1.8x. That makes sense too.

A15 is indeed much faster, with score of ~900 for 1.5GHz using Spec.

 

[ShintaiDK]

Intel will never get a huge chunk of the market if they not cutting the price of ATOM.

The cost of the CPU in an iPhone 5/Galaxy S3 of the retail price is what. 3-4%? Or around 10% of the BOM.

The CPU cost is hardly a factor if there is something better that makes your phones sell.

 

[IntelUser2000]

With tablets over $500 price of the SoC makes only a small part of the overall costs. But that is the high end and for that market you need the fastest SoC and the best package.

Asus's RT Tablet isn't lower priced either.

Intel's ASPs for last year on the Atom Nettop was $25 and Netbook was $27, which is in the level of ARM SoCs.

 

[sontin]

The CPU cost is hardly a factor if there is something better that makes your phones sell.

Not really. Right now people don't care about performance. It's the package of the product. That's the reason why OEMs looking for the cheapest SoC outside of Samsung and Apple. That's the reason why nVidia got nearly all Android tablets designs.

Clover Tail is not better than any "high-end" SoC on the market. OEMs will not chose Intel because of the name. That is not working in the Smartphone and Tablet market.

 

[ShintaiDK]

Not really. Right now people don't care about performance. It's the package of the product. That's the reason why OEMs looking for the cheapest SoC outside of Samsung and Apple. That's the reason why nVidia got nearly all Android tablets designs.

Clover Tail is not better than any "high-end" SoC on the market. OEMs will not chose Intel because of the name. That is not working in the Smartphone and Tablet market.

The products and development certainly dont agree with you. And everyone I know wish a faster smartphone, smart tv etc. Performance simply aint enough, not even close. And its something that makes people decide what to buy. Just look at Apple S models of phones for example.

 

[sontin]

Every SoC is getting faster: Tegra 4 will be faster than Tegra 3, Krait is faster than Snapdragon 3. SamsungEynox 5 serie is faster than Eynox 4.

But in the end outside of Samsung and Apple "more performance" means bigger chips and a higher price for OEMs. But the market is not rewarding more performance over the competition because image and mind share sells much better.

 

[ShintaiDK]

Every SoC is getting faster: Tegra 4 will be faster than Tegra 3, Krait is faster than Snapdragon 3. SamsungEynox 5 serie is faster than Eynox 4.

But in the end outside of Samsung and Apple "more performance" means bigger chips and a higher price for OEMs. But the market is not rewarding more performance over the competition because image and mind share sells much better.

Is that why the fastest smartphones are the most wanted? iPhone 5, Galaxy S3 and so on?

If the OEMs dont use the fastest chips they will lose sales.

 

[sontin]

Apple and Samsung. People want the latest from Apple and Samsung. And because of that Apple and Samsung can build much bigger Chips. They have the market size.

If Tegra 4 has a faster GPU than A6X it will not turn people from Apple to Asus. It would not even change anything because this market is not working like the discrete gpu market.

 

[pablo87]

Putting my thinking hat on, I feel Haswell and its successors, not Atom, is Intel's plan to takeover the smartphone/tablet business. I have 3 reasons to think this.

First, the power consumption vector seems to have little impetus to improve beyond where its at today (based on iPad 2 - YMMV). In linear programming terms, its not a goal but a constraint. Yet, if you look at every TICK and every TOCK from SB on, there is TDP improvement. Also the on-die integration of the chipset and the radio. So at some point in the not too distant future, Intel will meet this constraint with its higher end product.

Second, there is a want for better performance by buyers yet cost/die size isn't that big of a factor when only 3-4% of the BOM is the processor. In that scenario, Haswell/Broadwell/etc... trumps atom.

Third, Atom to me represents failure, SB/IB represent success. A big brand advantage.

One thing to be explored in this context is 7890M vs HD4000...

 

[Keysplayr]

Why do we want bigger chips again?

 

[blackened23]

The products and development certainly dont agree with you. And everyone I know wish a faster smartphone, smart tv etc. Performance simply aint enough, not even close. And its something that makes people decide what to buy. Just look at Apple S models of phones for example.

I really don't think the average consumer cares an awful lot about performance. Certainly if they notice slow screen re-draws and what not it may annoy them, but most consumers don't run benchmarks on their tablets and smartphones. They do care about user experience. If the performance is so bad that it hinders the experience -- such as slow screen refreshes - that will bother some people. But I don't think intel has such issues.

Unfortunately, it has taken MS too long to catch up on the user experience front for tablets - although I will say they're making good progress. I really feel like the Surface RT is a compelling product. I really love the idea - when I use tablets such as the iPad , I feel like they're a toy. Fun to use, but I can't do real work on it. The surface changes all of that.

I'm really excited about the Surface Pro. Full x86 compatibility? Hell yeah. If MS can nail the user experience, I don't think consumers will go crazy over open gl benchmarks. That said, i'm 100% certain that intel will not have performance issues with their mobile SOCs. I believe their goal is catch up on the battery life front, and once they do (they're really close now) they will be a strong player. With an ipad I can watch movies and download music while browsing the web all day. Doing real work? Pfft, not a chance in hell I would ever attempt that on an ipad. The native apps for productivity on the ipad are mostly a joke.

 

[Grooveriding]

Most consumers may not want the specific notion of performance as a hardware enthusiast does; looking to benchmarks to gauge the capability of hardware etc. What most consumers will want are the features better performing hardware will bring them in their mobile devices. They don't know or follow the internal hardware to know it is coming and what is possible, but when those features become available, they will want the better performing parts because of the features they'll enable in devices using them.

Intel has said they want into mobile. They have the technological edge on everyone, even Samsung, in their fabs. A few years and they will most likely be the dominant beast in mobile, and/or in tight competition with Samsung. Everyone else will be picking up the leftovers.

 

[Fx1]

I really don't think the average consumer cares an awful lot about performance. Certainly if they notice slow screen re-draws and what not it may annoy them, but most consumers don't run benchmarks on their tablets and smartphones. They do care about user experience. If the performance is so bad that it hinders the experience -- such as slow screen refreshes - that will bother some people. But I don't think intel has such issues.

Unfortunately, it has taken MS too long to catch up on the user experience front for tablets - although I will say they're making good progress. I really feel like the Surface RT is a compelling product. I really love the idea - when I use tablets such as the iPad , I feel like they're a toy. Fun to use, but I can't do real work on it. The surface changes all of that.

I'm really excited about the Surface Pro. Full x86 compatibility? Hell yeah. If MS can nail the user experience, I don't think consumers will go crazy over open gl benchmarks. That said, i'm 100% certain that intel will not have performance issues with their mobile SOCs. I believe their goal is catch up on the battery life front, and once they do (they're really close now) they will be a strong player. With an ipad I can watch movies and download music while browsing the web all day. Doing real work? Pfft, not a chance in hell I would ever attempt that on an ipad. The native apps for productivity on the ipad are mostly a joke.

If i ever see someone trying to do "real work" on a tablet then quite honestly ill laugh and point at them before feeling sorry for them.

about the only good thing a tablet is for work is for showing people pictures or taking tick box surveys. They are less useful than a pen and note pad.

 

[Fx1]

The cost of the CPU in an iPhone 5/Galaxy S3 of the retail price is what. 3-4%? Or around 10% of the BOM.

The CPU cost is hardly a factor if there is something better that makes your phones sell.

You clearly have no idea what your talking about. Samsung dont get "retail" revenue. How much do you really think they are making per phone?

16GB NAND costs about 2$ and 32GB costs about 4$ and 64GB costs about 30$

So ask yourself why doesnt EVERY phone maker just put 32GB inside? it would give them an advantage over the scroungy 16GB in the iPhone 5 or S3.

Every $ counts thats why. When selling millions of phones 2$ makes a huge difference.