Apple A8x

[Nothingness]

From a low-level perspective, sure, every companies is not perfect, but Intel is still the most successful semiconductor company, almost half a century since it was founded.

That is certainly true. But I was mainly disputing this claim:

When in history has Intel ever set a goal for its processors that it was not able to accomplish?

And I won't change my mind: historically Intel failed with their goal for their processors. This doesn't prove it will happen again, of course. But I wouldn't them to keep on succeeding, competition benefits everyone so as a consumer I don't want to see Intel alone from low to high-end 🙂

 

[Nothingness]

What do you take issue with?

I can only hope he wasn't talking about you quoting me, but you quoting Piednoel 😛

 

[witeken]

And I won't change my mind: historically Intel failed with their goal for their processors.

I see what you mean. Fair enough.

This doesn't prove it will happen again, of course. But I wouldn't them to keep on succeeding, competition benefits everyone so as a consumer I don't want to see Intel alone from low to high-end 🙂

Intel is a very good competitor. On the other hand, instead of degrading Snapdragon 600 to the low-end or making a dualcore SoC, Qualcomm is forcing OEMs to use the much slower A7 for their low-end offerings (or do they use it voluntarily?).

 

[Eug]

Well that's interesting. Here is the iOS 8.1 firmware for iPad5,4 which is the cellular version of the iPad Air 2. Tech reviewers apparently usually get the cellular versions, so they can test out their local carriers as well.

http://m.iclarified.com/entry/index.php?enid=8500

 

[Enigmoid]

Geekbench 3 SHA-1 and SHA-2 workloads operate on a single data buffer because we don't believe cryptographic hashes operating on multiple data buffers is representative of real-world usage.

Practically no Geekbench test is indicative of real world usage. Its all synthetic math tests.

 

[teejee]

Practically no Geekbench test is indicative of real world usage. Its all synthetic math tests.

And what is real world usage? If you know that Geekbench isn't real world usage then you must know what real world usage is.

In my opinion real world usage (CPU part) varies a lot for different users, but it is always a sum of different "synthetic" math tasks or other simple data manipulation.

 

[Enigmoid]

And what is real world usage? If you know that Geekbench isn't real world usage then you must know what real world usage is.

In my opinion real world usage (CPU part) varies a lot for different users, but it is always a sum of different "synthetic" math tasks or other simple data manipulation.

I agree completely on the second part.

Personally if I could change geekbench.

Compiler support targeting all relevant extensions for ARM, x86, etc. Less encryption/decompression (Do you really need JPEG and PNG tests? Can't you simply put that under "media compression"). Memory latency tests (just as important as bandwidth). Virtual machine (JS) performance. Character recognition.

The point is if you are going to use "synthetic" math tests then exploit them to the fullest. Geekbench seems to say "We will use a bunch of synthetic math tests to evaluate real world performance but wait, don't make it too synthetic, after all, these completely synthetic math test must somehow apply to the real world".

Now I understand your dilemma, these tests need to run on a variety of platforms so they can't for instance use 4 GB of RAM or something like that.

 

[Mopetar]

To some degree synthetic tests are as good as it gets. Real-word performance can't always be accurately measured. Sure you can devise a test to measure real-world performance for something like web-browsing, but then you end up depending on the OS and the browser, and while that might be okay to compare Android phones, it's essentially useless for cross platform comparisons. The other problem with real-world tests is that no one can agree on what is a good test. There's always going to be someone who will say the test is useless because it isn't even close to their use-case.

Synthetic tests that are close to the metal might not be the most useful barometers of real-world performance, but they're at least the most fair in terms of the fact that they measure what they measure, even if might not be all that useful in terms of day-to-day typical usage patterns.

 

[jfpoole]

Practically no Geekbench test is indicative of real world usage. Its all synthetic math tests.

Saying that no Geekbench test is indicative of real-world usage is not fair or reasonable. While some of the tests in Geekbench 3 are synthetic tests (Mandelbrot and STREAM being the prime examples) most of the tests either use libraries (e.g., JPEG, Lua) or algorithms (e.g., Dijkstra, FFT) found in real-world applications.

 

[Dribble]

You logic does not make sense; Intel obviously knows that, yet they decided to enter the market anyway, which costs them billions per year.

You have a point, though. If Paul Otellini knew what would happen, Intel was now making tons of money. Now they'll make less because they have to compete with other companies that already participated in a race to the bottom. Qualcomm won and Qualcomm is doing absolutely fine. But that doesn't matter. Intel is losing a billion every quarter. They aren't going to give up now, they are simply going to take Qualcomm's place and become profitable. Once they have the position, who knows, maybe they'll be able to ask higher prices.

Qualcomm isn't the cheap end, lots of chinese soc makers beat them. Why is Intel in mobile? - because it's the future. They have no choice, but unlike the PC market they don't control it.

Intel can also compete. They can outcompete, until their fabs are so far ahead and their technology is so much better than Apple, then Apple won't have any other choice if they don't want to fall behind the leading edge of performance

But most don't care about *leading edge* performance. Most phones sold don't have a cutting edge soc in them. Intel do have a fab edge but it doesn't make that much difference.

x86 is not less efficient in any meaningful way. This was debunked by AnandTech almost 2 year ago and is now especially being proven by Core M.

Which doesn't fit in a phone. They still haven't made a phone soc. Now they can just about do a tablet one. This is with all their huge R+D and process node advantages. ARM probably pays it's whole dev team less then a few Intel execs get in bonuses and they pump out designs for soc's that work great in phones which some Chinese company produces for peanuts. Either Intel R+D is bunch of monkeys compared to the little ARM dev team, or x86 is a huge drag?

Fear, uncertainty, doubt.

(Or do you have anything to back up your statements?)

Well there's the fact that Apple and others are showing clear intentions not to use Intel going forward even when Intel must be offering them a very good deal - they are looking to remove Intel chips from their machines. It won't be long till the mac-book air stops using whatever the latest Intel chip is and starts using one of Apples own designs.

 

[witeken]

Q
But most don't care about *leading edge* performance. Most phones sold don't have a cutting edge soc in them. Intel do have a fab edge but it doesn't make that much difference.

It isn't too difficult to comprehend: if 2 companies have an SoC for the same price, which one would they chose? The one with the best feature set, performance, power.

Which doesn't fit in a phone. They still haven't made a phone soc. Now they can just about do a tablet one. This is with all their huge R+D and process node advantages. ARM probably pays it's whole dev team less then a few Intel execs get in bonuses and they pump out designs for soc's that work great in phones which some Chinese company produces for peanuts. Either Intel R+D is bunch of monkeys compared to the little ARM dev team, or x86 is a huge drag?

Why do you think you can't make Broadwell for phones? Thermals? Just reduce clock speed and it will still be a lot faster than anything else. It's because there is no such thing as a productivity phone market, which Microsoft addresses with its Surface Pro devices. You make a tablet with Core M and you can attach a keyboard to it.

There might be some other limitations but that's simply because it was made for tablets.

Well there's the fact that Apple and others are showing clear intentions not to use Intel going forward even when Intel must be offering them a very good deal - they are looking to remove Intel chips from their machines. It won't be long till the mac-book air stops using whatever the latest Intel chip is and starts using one of Apples own designs.

You have not a single credible rumor to back those "clear intentions" up.

 

[Arachnotronic]

Why do you think you can't make Broadwell for phones? Thermals? Just reduce clock speed and it will still be a lot faster than anything else. It's because there is no such thing as a productivity phone market, which Microsoft addresses with its Surface Pro devices. You make a tablet with Core M and you can attach a keyboard to it.

Broadwell-Y lacks a lot of necessary IP blocks to be competitive in a phone. Further, I would bet you that the BoM that a Broadwell-Y drives is simply too high/too large to be able to fit into a phone.

To illustrate my point, compare the block diagrams of a Core M with a Moorefield:

Notice how Moorefield has an integrated image signal processor, a video signal processor, and just the right I/O for a phone?

It's also a single die, likely in a much more compact chip package. It also supports PoP memory, which is something Core M doesn't.

No, Core M would be a dreadful phone chip.

 

[Roland00Address]

Why do you think you can't make Broadwell for phones? Thermals? Just reduce clock speed and it will still be a lot faster than anything else. It's because there is no such thing as a productivity phone market, which Microsoft addresses with its Surface Pro devices. You make a tablet with Core M and you can attach a keyboard to it.

Sorry but this image http://images.anandtech.com/doci/8515/DSC_7126_575px.jpg

Means Core M in its current 14nm Broadwell generation is too big for phones. The PCB itself is bigger than most phones. Within time I foresee Core M ending up in high end phones but not even close to being possible within the next 12 months with current technology or a phone so large it is a tablet and not really a phone.

 

[Arachnotronic]

Sorry but this image http://images.anandtech.com/doci/8515/DSC_7126_575px.jpg

Means Core M in its current 14nm Broadwell generation is too big for phones. The PCB itself is bigger than most phones. Within time I foresee Core M ending up in high end phones but not even close to being possible within the next 12 months with current technology or a phone so large it is a tablet and not really a phone.

Also a solid argument.

 

[witeken]

Broadwell-Y lacks a lot of necessary IP blocks to be competitive in a phone. Further, I would bet you that the BoM that a Broadwell-Y drives is simply too high/too large to be able to fit into a phone.

To illustrate my point, compare the block diagrams of a Core M with a Moorefield:

[imgs]

Notice how Moorefield has an integrated image signal processor, a video signal processor, and just the right I/O for a phone?

It's also a single die, likely in a much more compact chip package. It also supports PoP memory, which is something Core M doesn't.

No, Core M would be a dreadful phone chip.

No duh. Core M isn't made to go in phones, that was sort of my whole point. But Intel could 'easily' take the Core IP and replace the Airmont CPU with it. There's no reason I know that Core couldn't go in phones. They can simply reduce the frequency by 100MHz like Apple did.

 

[Roland00Address]

No duh. Core M isn't made to go in phones, that was sort of my whole point. But Intel could 'easily' take the Core IP and replace the Airmont CPU with it. There's no reason I know that Core couldn't go in phones. They can simply reduce the frequency by 100MHz like Apple did.

I 100% agree with what you are saying Intel could in theory could take broadwell core m IP and create a new separate die with core m IP in phones. In theory it could happen.

But in reality it is not going to happen in the next 12 months. Since it is not going to happen in the next 12 months it will not be broadwell when this does eventually happens. The earliest it could happen would be at least 1 generation (skylake), but far more likely it will be at least 2 generations (cannonlake) or even more (cannonlake successor has not been announced yet).

That said I am extremely excited for broadwell's core m shows that it is completely feasible on a practical engineering level to do so. Soon we can get what we now consider ultrabook performance in something small enough to be a 5" phone or a phablet 5.5 to 6.5." It may not be this year but it is possible to do so in the next 3 years if Intel wants to do so.

 

[witeken]

I'd say Intel could do it with Skylake, but they'll probably focus on 7-8", with Core for Smartphone coming with Cannonlake, if it ever happens.

 

[sm625]

When in history has Intel ever set a goal for its processors that it was not able to accomplish?

They failed in the $99 x86 tablet claim. The only $99 x86 tablets we have are bargain bin throwaways from companies who tried and failed and are forced to fire sale their product. We'll never have $99 x86 tablets as long as intel is charging a 60% margin, and intel will never compete in mobile with 60% margins either. There are half a dozen companies out there who are willing and happy to accept margins as low as 15%. Intel cant compete with that.

 

[Arachnotronic]

but far more likely it will be at least 2 generations (cannonlake) or even more (cannonlake successor has not been announced yet).
[/i]

Icelake.

 

[witeken]

They failed in the $99 x86 tablet claim. The only $99 x86 tablets we have are bargain bin throwaways from companies who tried and failed and are forced to fire sale their product. We'll never have $99 x86 tablets as long as intel is charging a 60% margin, and intel will never compete in mobile with 60% margins either. There are half a dozen companies out there who are willing and happy to accept margins as low as 15%. Intel cant compete with that.

At the same time, there are still people saying Intel is giving its full Bay Trail chip away for free. There will probably be more cheap tablets when SoFIA launches.

 

[Arachnotronic]

There are half a dozen companies out there who are willing and happy to accept margins as low as 15%. Intel cant compete with that.

Let's do some math.

It costs about $300 million to develop an SoC in sunk, fixed R&D costs. If you want to be really, really generous, let's say it's $200 million.

So, in order to *break even* on this sunk R&D cost, a chip needs to generate at least $200 million in *gross profit* which is revenue less cost of goods sold.

Now, if a company were to sell a chip for 15% gross margin, then how much revenue would they need to generate in order to simply break even?

$1.33 billion.

Now, recognize that the entire tablet processor market (according to Strategy Analytics; Source: http://www.strategyanalytics.com/default.aspx?mod=pressreleaseviewer&a0=5480) was about $3.6 billion.

So you're telling me that there's room for "half a dozen" companies, each selling their chips at 15% gross margin, to capture a $3.6 billion total market?

 

[Eug]

I'd buy a Core M ultra-ultraportable. Phone? Not so much.

Call me when they actually release a Core derivative suitable for a phone, and in a phone I actually I might be interested in. Until then, I don't care.

 

[ViRGE]

I'd buy a Core M ultra-ultraportable. Phone? Not so much.

Call me when they actually release a Core derivative suitable for a phone, and in a phone I actually I might be interested in. Until then, I don't care.

Considering Intel's desired margins on Core, it's unlikely you could afford such a device.🙁

 

[witeken]

Considering Intel's desired margins on Core, it's unlikely you could afford such a device.🙁

It all depends on how much of a margin hit Apple (or...) wants to take. A $1000 iPhone with 256GB, Cannonlake and Gen10 should be easily possible with sane margins.

 

[pw257008]

It all depends on how much of a margin hit Apple (or...) wants to take. A $1000 iPhone with 256GB, Cannonlake and Gen10 should be easily possible with sane margins.

But a $700 with 16/32gb?