Intel looks to build ultra-efficient mobile chips Apple 'can't ignore' - Page 3

Unfortunately that won't be out until the 1st half of 2013 (potentially more than a year away) and the results don't look good enough to usurp ARM's throne. I'm sure Intel will get some vendors on-board, it might be an ideal fit for Win8 tablets, and it's certainly a lot closer than they were to ARM a year ago but will be good enough?

Another issue I see is the complexity of the ASIC which could be a deterrent for companies like Apple. Apple licenses the ARM reference designs and then build's their own ASIC according to some general specs thus making it a unique and ideal chip that suits their specific needs. I don't think they can technically call it Cortex with the number of low-level changes they make. Will Intel offer them the same options as ARM? Not from what I've seen.

PS: Your link doesn't work because the period at the end of the sentence is being attributed to the link.

No, Apple needs to continue to outperform by:

1) better GPUs than the competition.

2) stay at dual core and improve linear performance. For example ARM Cotex A15 designs, smaller topographies and faster GHz.

Tablets are not well suited for more than 2 cores.

I don't think anyone here knows that. iOS is based on OS X - which handles multiple cores and threads very nicely. I suspect that an iOS based tablet would easily handle 4 or more cores.

As for the rest, unless you're part of Apple's design team, you don't have any idea what they have planned. Frankly, I don't care if there is one core or a thousand. I don't care if it uses faster GHz or not. I don't care if it's A15 or A2000. All I care about is performance. As long as Apple continues to provide competitive (or superior) performance, they've done their job.

 

Do you mean now or do you mean in general and forever? Can you elaborate on that?

Cute

All you have to do is look at the use cases of a tablet. They are designed, in Apple's implementation, to be immersive. The screen is dedicated to a single app view that may have multiple sub views. This is iOS developers guide 101.

While there are gains in performance to be had on multiple cores, you start getting limited after 2 due to the single threaded nature of most processing tasks. So yes, you may have hundreds of theads running but many of those are queued up and are serial in nature. In short, you quickly reach a limit on how much a tablet or phone benefits from multiple cores. Most people hit that limit at 3-4 cores on a desktop. Tablets, I am betting, are around 2.

So if you have limited silicon to work with, where do you put your resources? Given the Retina screens, the GPU is the first obvious choice. Pushing pixels is a highly parallel task and doubling the performance of the GPU doubles the performance of times when pixels are being moved; that happens all the time. From games to word processing, iOS benefits directly from faster GPUs.

The other option is to increase clock rate and increase single thread performance. ARM, really lags traditional designs, like the Intel Core micro architecture, in this area. To get there you use smaller topographies (allowing higher clock rates at the same power usage) and updated ARM designs like the Cortex A15.

In the future, like iOS 8.0 or 9.0, the design paradigm may shift to a multi windowed touch iOS but that is still a few years off and hardware should be optimized for the current use cases and not some far off future use case that may never happen.

You keep saying that, but haven't substantiated it with anything but your opinion.

That's my point. You don't have any evidence to support your claim that a tablet can't use more than 2 cores. Please explain why Apple put 4 graphics cores in the latest iPad, for starters.

(Yes, I know that graphics is different than CPU, but there's no reason that the OS and apps couldn't be multithreaded enough to use multiple CPU cores, as well as graphics cores).

There's a little more information on their new Atom SoCs here: http://semiaccurate.com/2012/05/11/intel-outs-merrifield-and-6331/

 

Fair enough, I look forward to it. Instinctively I'm sceptical, and Intel has a very narrow time window (2012-2015) to really pull it off. Like I said, if they can, then it will be a massive success story for tech in the early decades of this century. Thanks for pointing the article out, I don't usually spend time on Anandtech but I'll check it out.

 

Personally I like the roadmap that ARM proposes, ie. big.Little - two beefy cores for extra performance paired with two lean cores for most tasks. Throw in 4 or 8 core next-gen mobile GPUs, and it looks like a winner to me, at least for iOS.

 

A legitimate point. But also keep in mind Apple doesn't make CPUs for a living, Intel does :)

 

 

 

Some concerns from http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone/2

 

"In addition, if the platform gets too hot, the display brightness will be clamped to 50%"

 

"For one, Flash 11 isn’t available for the X900, and throws an error in the market. The device does come preloaded running Flash 10.3 however, which gets the job done although is a bit dated. In addition, although Netflix would download, the installer would throw a ‘package file invalid’ error upon install. This is what leads me to think there’s some APK interception in the cloud and perhaps translation up there, and Netflix DRM not translating, but that’s speculation. Other than this, everything else I encountered works flawlessly, I wager your average Android user wouldn't be able to tell that this is running on a completely different architecture."

 

So for 2012 it looks like x86 Android is an interesting proposition, and could eat up Android share of ARM. But nothing Apple "can't ignore" this year at least.

 

I remain of the opinion, Intel might be able to pull it off in 2013-2015, whereby Apple using Intel in iOS would be the vindication for Intel.

 

Given ARM's roadmap as well and Apple's custom ARM designs, it's a fairly lofty goal for Intel.

 

I admit thanks to the posters here, I am less dismissive about x86 phones and tablets, but remain skeptical.

 

It is interesting to note as well than Intel has not only got to climb uphill here but carry the bag of rocks that is Android, and the broken horse-cart that is Windows 8. I'm not saying there's no place for Android and Windows8, but delivering the user experience Intel wants, yet being dependent on Android and Windows 8, is a burden in several ways. Not to mention Android and Windows 8 will be seen as the non-Apple "more affordable" market, so anyone using Intel chips will want those chips at low, low prices... which Intel will do to the point of, or beyond, illegality. But that would let Intel corner the bottom-end of the smartphone and tablet market, which is not a great place to be. Tough times ahead, luckily Intel has the money in the bank with Sandy Bridge and Ivy Bridge.

 

Still, it would be a whole bunch less of a headache if Intel just fabbed ARM for elite clients.

Edited by nvidia2008 - 5/14/12 at 11:17pm

Thank you for proving my point. Apple included a much faster GPU in the current iPad. That was bullet #1. They did that by using a quad core GPU not a quad core CPU.

Likewise, what I stated was not just opinion, it was very well reasoned opinion based on 25 years in the industry. Likewise, my statement of Apple's view of the iPad is actually covered in WWDC sessions I have been to and are available on line.

Go learn about software design and then we can communicate better.

From what I remember, this is a single core chip with a fairly weak GPU. Given that it's compared to dual core models, that makes its battery performance somewhat worse than it appears in the charts. By the time Intel has a dual core model out, the other newer chips will also have better performance, and better battery life. They ate all going to new process tech, so Intel won't have an advantage there.

That test isn't needed. The main reason why Apple wants its own IP in its chips is to be able to leverage its OS together with it. So even if we could get some version of Android to work well enough to matter, it still wouldn't be a fair test. That's because Apple likely has microcode and firmware differences, as well as things such as the known photo processing DSP that's unique to them on the chip. Who knows what else is on there? There are chunks of areas on their chips that no one can figure out. They must be doing something that Apple needs.

Any Android install won't be able to take advantage of all that, and indeed, it's very possible that Apple eliminated portions of the circuits they don't use, for the reason of efficiency.

So what would be learn?

Let's be very careful when talking about "on time" delivery of Ivy Bridge. It was expected, by Intel, to be available for production in the first quarter. It's now late middle second, and only a few chips have trickled out. We won't see all models out until the second half of the year.

Originally, Intel expected 22nm to be out forth quarter 2011. But everything's slipped. Some of that may be due to AMD's general incompetence. A little bit to the slowdown of x86 computer purchasing the past few quarters. But the majority of it is due to the tuning of the production process.

I fully expect Intel to say, this early, that 14nm is doing well, and is on schedule. But considering that production plants are still being built, and very little 14nm production equipment has been built, much less delivered, by the manufacturers, I don't see how anyone can take Intel's word at this so early in the cycle. The articles I read in my journals say that this is going to take more time than expected originally.

As Intel has come out with new nodes at the end of the year, it's certainly not unreasonable that the slide to the middle of the next year can be interpreted as a lengthening of the cycle.

And if they are saying May now, that will slide as well, possibly to the end of the year, which will bring it to 2015, as industry experts are expecting. At any rate, full production ramp-up won't be until some time in 2015.

That seems shortsighted. Everything evolves. If we haven't learned that from using computers for decades, then we haven't learned anything. IOS is in a constant flux of evolution. Apple adds API's, often by the hundreds, every year. These cover many areas. They went from one core to two. What makes you think that their vast experience of having multi core, and chip machines isn't being taken advantage of here? I would think that it is.

Any single threaded area can be reworked as a multi threaded process if it's practical. Apple does take their time, but they do get there. There are advantages to quad core chips. Video editing is something Apple was interested for a tablet going back several years before the iPad came out, as evidenced by a patent showing that editing as an application. There are few things that benefit by multi cores as much as video editing. You can bet that Apple wants to speed that up.

As for your immersivness comment; the more natural something appears to be, the more immersive it is. Therefor, the faster the system is, and the better the quality of the screen and related touch response, the more natural it will feel, and the more immersive. If you don't have to wait for a response to something, then it will feel more natural, and without that time lag, it will be more immersive.

I don't know how old you are, but I remember when, at a computer show here in NYC, during the mid 1970's, when there were over 100 different small computer companies making 8 bit machines almost all with their own OS's, I was invited by one of those companies to see a demonstration of their new line of machines. We sat down in front of the green screen monitor, and typed. "Isn't that amazing" I was told by the guy showing me the machine, "the letters appear on the screen almost as fast as you type!"

We expect more these days, of course. But there is never a limit to the speed of a machine. It will never make us happy until everything we do will be instantaneous.

I have no doubt that we will see a quad core SoC for the iPad, at least. When is another question. It may or may not come next year, but it will come.

The limits in the performance of the new iPad, which I'm typing on now, are due to the cpu portion of the SoC, and memory bandwidth. We can see that in games and some 3D apps, for example.

At the least, we'll need that dual core Cortex 15 chip. If Apple improves the graphics performance next year with Imagination series 6000 GPU's, they will need a lot more power in the CPU to drive that graphics. Some things are gpu limited, and some are CPU limited.

A four core Cortex 15 will likely be ready for the one after next year. Maybe even for next year.

You know as well as I do that this statement is categorically false. Many prosecutes in any program are serial in nature and 1000 cores will not make them run faster. 95% of the tasks I do on the iPad are single thread and GPU limited. From emails to games. Having 4 cores would NOT double their speed.

More cores will NOT improve memory throughput. You know that.

Having 4 cores will not allow you to controll a Rougue series 6 GPU any faster than having 2 cores will. Having 2 1.5 GHz Cortex A15 cores will allow you control a Rouge Series 6 GPU faster than having 4 1GHz Cortex A9 cores.

So where do you put your effort in the short term?

1) Faster GPUs.

2) staying Dual Core.

Where has apple put their effort?

1) Faster GPUs.

2) Staying dual core.

OMG!!! What a shock. PS: I remember using punch cards to program the campus mainframe. Sure, it would be great to have an 8 core 4GHz Cortex A20 with a hyper Rougue 7000 all on 11nm tech. We just are not there yet.

That's why I said if it's practical. Memory needs to be speeded up as well. I mentioned that in another post. I've also mentioned what they're doing now, and I said it was appropriate. But for the future, they will need to go to 4 cores. You're bringing up 8 cores, and that's well into the furure. But don't say it will never be wanted. I have a Mac Pro with 8 cores, and 16 virtual cores, and on a fair amount of work, most, and sometimes even all of those cores are in operation.

Actually, wouldn't it depend on the algorithm in question? Not all tasks can be parallelized. Amdahl's law bites.

First a disclaimer, I am doing research in parallel computing, I could tell you more but I'm bounded by NDAs. The keyword is in-parallel, if your code is sequential and can't be parallelized and believe me a lot of code can't be easily parallelized then you hit diminishing returns right around 2~4 cores.

Second, here ya go:
http://en.wikipedia.org/wiki/Amdahl%27s_law
See that chart at the top right corner of the article, it says it all.