Intel phone boss: 'Multi-core detrimental to Android mobes' Intel’s head of mobile has dissed handset-makers that have already adopted multi-core processor architectures, saying that most implementations so far are actually “detrimental”. Mike Bell, general manager of Intel’s Mobile and Communications Group, also said that smartphone users who buy into Chipzilla's platform can expect …
Multicore is great a match for Android on ARM but there are no magic bullets for scheduling and I/O. The mobile phone market has validated both the multi, specialised core (CPU for managing the user, DSP for signal processing, etc.) and SoC (squeeze processing and radios onto the same component), which is why Intel and AMD are playing catchup with the unified architectures.
ARMs smaller, weaker cores offer greater efficiency for the majority of tasks, whereas as Intel's larger, stronger single core shines at individual disciplines. If it weren't for the GPU Intel would shit on ARM for rendering websites which is a surprisingly complex task. But the GPUs are there offload the rendering to.
Intel's advantage in chip-manufacturing is essential to be able to compete in the mobile market. But, as long as its unit costs stay at least a factor higher than the competition, it can't afford to maintain that advantage forever. And it's not even as if the competition is that far behind with the move to 32nm already happening.
The only way Intel's lead in chip-manufacturing would be an advantage is if they get an ARM license license and make ARM processors. The Atom is so far behind that their fab-advantage is completely wiped out.
They have also made a massive mistake in putting an ARM emulator in so they can run Android apps containing native code without the developer adding an x86 build to their app. So guess what, developers won't bother recompiling an x86 version because all their apps will already work on intel phones, but they will run slower making Intel look even worse than they already are.
If Intel swallowed their pride and started manufacturing ARM cores using 22nm 3d transistors they might actually dominate the mobile market.
-- "Intel's advantage in chip-manufacturing is essential to be able to compete in the mobile market."
This word "essential": I don't think it means what you think it means. Intel isn't even present in the mobile market to a significant degree, so no advantage they have could possibly be considered "essential to compete in the mobile market."
Perhaps if Intel has a closer look at their advantages and strategies to see what is failing them, they will find what they need to compete. The key is likely in the latter, not the former.
Snake Oils Sales 101: Always divert the attention from the real defects in your product and make the story about some advantage - even if you have to make one up.
Nobody cares about single vs multi-core efficiency. All we care about is whether the damn device is responsive enough. With time Android and other OSs will become multi-core friendlier making his point moot in the next year or so.
Intel still suck at battery life. That is much more important.
ARM is multi-vendor. That is much more important. That allows for far better innovation as anyone can make new cool SoCs with ARM cores. Intel is single vendor: you only get the SoCs Intel chooses to make. No competition and the industry stagnates.
Imagine if Intel had cornered the phone market in the 1990s. We'd still be walking around with brick phones.
It's a little bit more subtle than that. What he's really saying is 'If you losers want to catch Apple you'll need to partner with us 'cause we're the ones who really know what we're doing. The only reason we haven't shat all over the competition up 'till now is that we're too busy printing money out of PC's, but just you wait'
“no one has a market advantage right now”
So, that almost universal dominance of the arena by ARM processors isn't a market advantage? Maybe he means that no single manufacturer of ARM chippery has an advantage over the others?
That's a good thing, it's called competition. I know it's an alien concept to Intel, but they should probably have looked it up before wading into this field.
ARM licences the core design to anyone that wants to use it. So as a result you see SOC designs with varying built in hardware and an ARM core. Lots of choice and competition since it is the complete chips that matter, not what their execution core is.
With Intel you get what Intel wants you to have. Hardly useful for compact efficient designs with minimal component counts.
"We got no SoC licencees yet, and we only got one chip family. Please buy our SoC not Qualcomm's or whoever.
We got no Wintel leverage in this market. Please buy our SoC..
We got nothing useful at all really in this market, but please buy our SoCs, we really really need a success outside the Wintel market, because we haven't had one for over a decade."
Not quite what he said, but at least as factual, no?
I'm looking forward to the first teardown from iSuppli or whoever. It should be interesting reading.
This post has been deleted by its author
Bell [added] that having “access to source code where you can go in and really work with it” is a massive advantage.
Bell was more tight-lipped on the open source dogmatic aspects of working with Android. Asked if it was feeding back its innovations on the platform, he said that it was doing so "where it [was] required", but added: “[I] don’t like doing R&D for my competitors.”
“In some of the use cases we’ve seen, [the] second core is detrimental because of scheduling.” Having looked at the multi-core options on the market, he said, the performance didn’t justify “the size and cost of putting in that part".
I'm sorry... what? The Android kernel is Linux, and since kernel 2.6.26, Linux has had an awesome multicore thread-scheduling algorithm. Maybe he's saying the Dalvik VM doesn't take good advantage of it? Or is he just slagging 50 years of continuous improvement in the state of the art of symmetric multiprocessing?
I guess it's more like post #2 (@Anonymous Coward) says: "We don't do it, and our competitors do, so it must suck."
BTW, I'm pretty pleased with the performance of my Motorola Droid 4. If multi-core is a performance detriment, I haven't noticed it yet.
It's not though. With Tegra 3 we are headed into Asymmetric MultiProcessing (AMP) on our way to Heterogeneous MultiCore territory. It's going to take a while for Intel's august code veterans to convert to this alien new religion.
They're best quit of their Windows habit, as it is done with them.
Last I checked, June was the 6th month of the year, and the 3rd (and hence last) month of the second quarter, so wouldn't the next quarter be the one going from July to September? Clearly a quarter has not just begun if it is June. In fact a quarter is about to end. Someone needs to check their calendar, and it isn't Intel.
Especially if you keep making them faster until they show up in tablets and (horrors) laptops!
What's with this process migration across cores and low power core option? Please stop innovating!
As Apple will tell you, the consumer market has remarkable price insensitivity for things they really like.
Intel are claiming they have form in software? Don't make me laugh! I am currently evaluating their VTune and inspector tools. They are a total crock. Valgrind is about 10X better and more useable than their checker tool, their profiler is ok (when it doesn't crash, which is once a day), but no better than gprof or Google's profiler for Android. I certainly won't be paying for any of their shitty tools. Its more cost effective to port my software to linux and valgrind it.
On the other hand, I have also evaluated ARM's compiler and profiler tools: they are totally awesome and worth every penny of their fat price tag (which I can't afford).
Also their phone looks ugly and will be ruined by Orange's pre-installed crapware.
For balance, Intel's compiler is ok and they make good desktop processors.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
