Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intels-mind-boggling-process-advantage-explained.7871/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Intel's Mind-boggling Process Advantage Explained

benb

Well-known member
Specification summary:
-Apple A9X vs. Intel Core m3-6y30 processors are roughly comparable
-MHZ: 2260 (Apple), 900-2200 (Intel)
-Die Sq mm: 99 (Intel), 147 (Apple)
-Graphics cores: 24 (Intel), 12 (Apple)
-Graphics performance (PCMark Icestorm Unlimited): 33574 (Apple), 40995 (Intel)
-Cost: Apple iPad Pro ($799), Microsoft Surface 4 Pro ($899)
-Process nodes: Intel 14nm, TSMC "16nm"

Why "Mind Boggling": Intel doesn't advertise their die size, transistor count or core counts publicly anymore, so as to throw competitors off. However, these things are relatively easy to determine and public. Chipworks, Anandtech and other sources publish these things. They also publish the configuration for the Core m3 as "2+2", another obscuration technique. It means 2 cores and "GT2" graphics performance. GT2 refers to 1 slice, 3 subslices, totalling 24 execution units. This then rolls into a HD 515 designation for 24 execution units with a maximum clock speed of 850 MHZ.

This is a lot of hiding behind jargon and numbers, but it is all public now. Finally, the die map for the Core i7-6700K has been published, showing, at the left, how those 24 execution units are arranged, with 16 in the upper left side, in line with 2 of the 4 cores, and 8 more in the middle-lower left corner in line with cores 3 and 4.

77a.jpg



This is what is mind-boggling about Intel's process advantage: In order to include 24 EUs in the Core m3, they must include Cores 3 and 4 in the bargain, but disabled. That is 20% of the already superior (147 vs. 99 sqmm) die area. If they had designed a purpose-built high-end tablet chip, like the A9X, they could have probably reduced the die size to 70 or 80 mm. Intel is simply killing it in the high-end tablet chip sockets. And I think Apple may have to look hard at giving Intel a shot at it's next "X" chip, simply because they can get a better deal; better performance and lower cost.
 
By the time the next X chip is released TSMC 10nm should be available so this is not a realistic comparison.
 
By the time the next X chip is released TSMC 10nm should be available so this is not a realistic comparison.

It is my hope that the Apple A10X will in fact be TSMC 10nm. Now that would be an interesting comparison. If not we will need to wait until the 2018 Apple iPad refresh (A11x), which would be less interesting. Unless the A11x is TSMC 7nm of course which is certainly possible.

There was just a 6.1 earthquake in Taiwan. I'm from California so a little shaking doesn't scare me but it certainly was a fun ride on the 18th floor of the Hotel Royal!
 
Sorry, maybe it is caused by my bad english, but i don't fully understand what you wanted to say.

Anyway, here is illustration of A9X die: https://g.foolcdn.com/editorial/images/186872/a9x-annotated_large.png

Area marked as GPU/CPU core clusters is only around 33%.

3D mark results is interesting, but i do not think, that it is unbiased. You can't compare even different architectures of that same manufacturer in this benchmark (or example AMD Tahiti and Tonga) so what should i think about comparing different platforms?

But look at raw performance: 409,6 GFlops at 400 MHz vs 422,4 GFlops at 1100 MHz for Intel. So here will decide better optimization of application in which we test.

And main problem is TDP. We don't know real numbers but Intel need active cooling, what is pretty bad in mobile/tablet world.
 
D.A.N., TSM stock price has spiked from ~23 to ~25 in the last several days, since the conference you're attending started. Any word on why that may be?

Folks, II wanted to bring together some public information that chipmakers, Intel and Apple, have been trying to hide or obscure. It is clear that in the 4.5 W, fanless socket, Apple has "a ways to go" before they can beat Intel m3. In graphics, in actual devices, in a direct comparison, PCMark Icestorm Unlimited. In processing capability, there are many comparisons, none which I like very much; however the overall take-away is Apple winning in a few, but m3 can beat A9X in most. Finally, the cost of the devices using the chips are similar ($799 vs. $899) although Apple has the advantage here. I think eventually we will see the m3 chips in $499 tablets though. And Intel's device is much smaller today (99mm) vs. 147mm, 32% smaller; it could realistically be smaller still and still perform as well, based on the wasteful die layout they went with, mainly to support larger PC-class chips.

This board likes to hear from diverse viewpoints, I hope. There is a belief that ARM-based chips can scale up in performance and win against x86. I don't see the evidence for this, and pretty convincing evidence of the opposite exists today.
 
Last edited:
I don't think there is much debate that, at the current snapshot in time, Intel has the process and performance advantage but not the cost/marketshare advantage. I do think the ARM ecosystem may eventually catch or possibly even surpass Intel in performance as foundry processes catch up, which could happen at 10nm or 7nm.
 
@Jozo035,
From the A9X die photo, it looks as if Apple pared the SoC back to just the CPUs, GPUs and associated memory and control, it would be well below 99mm^2. Benb, the real question is whether the non-GPU and CPU silicon in the A9X (and A9) adds addition value that doesn't show up in the standard benchmarks, only in iOS.
 
@Jozo035,
From the A9X die photo, it looks as if Apple pared the SoC back to just the CPUs, GPUs and associated memory and control, it would be well below 99mm^2. Benb, the real question is whether the non-GPU and CPU silicon in the A9X (and A9) adds addition value that doesn't show up in the standard benchmarks, only in iOS.

I agree but just want to point out that Intel processors nowadays also include blocks beyond CPU, GPU and memory. It would be indeed interesting to know what the extra features on Apple SOC are and how much "real estate" they take.
 
@lilo77,
The one known block in the "extra feature" real estate is the M9 motion/sound processor. There's also one other caveat on the Skylake Core M3-Y - the package actually contains two die, the processor SoC, plus a PCH hub in 22nm silicon, so it's a bit unfair to compare the standalone A9X with just the CPU/GPU die on the Core M3, when it takes another big chunk of silicon to do the whole job with IO. I don't know the exact size of the PCH die, but can see it here:
Processor Lists and Conclusions - The Intel Skylake Mobile and Desktop Launch, with Architecture Analysis
 
Really hard to compare how many transistors Apple needs to run iOS with how many transistors Intel needs to run several generations of Windows. I'm sure if Microsoft weren't in this conversation, the Intel footprint would look a lot slimmer ... oh, wait, that'd be Broxton.

There's also a capacity problem, one TSMC and Samsung are very familiar with in servicing Apple. Theoretically, Intel still has Fab 42 sitting empty (not "mothballed", but empty). I can't see Intel making a cap ex that big to bring Fab 42 online just to support Apple, and they aren't going to swing Skylake capacity out of PC/tablet/2-in-1 biz at other fabs.

Intel's model always has been service 10 big customers and send the rest through distribution. They don't do well with customizing production for one customer (see: Qualcomm), and the foundry biz is all about customization. If I were Intel, I'd take the same investment it would take to win Apple mobile at this point and go after the IoT - exactly what they are doing.
 
Specification summary:
-Apple A9X vs. Intel Core m3-6y30 processors are roughly comparable
-MHZ: 2260 (Apple), 900-2200 (Intel)
-Die Sq mm: 99 (Intel), 147 (Apple)
-Graphics cores: 24 (Intel), 12 (Apple)
-Graphics performance (PCMark Icestorm Unlimited): 33574 (Apple), 40995 (Intel)
-Cost: Apple iPad Pro ($799), Microsoft Surface 4 Pro ($899)
-Process nodes: Intel 14nm, TSMC "16nm"

If we put the unit price of each processor into consideration, I have to ask why Intel's price is so much higher? Is it because Intel needs the high profit margin in order to satisfy stock analysts or Intel's cost structure is just too high?

Apple A9X: $37.30
Intel Core m3-6y30: $281

Even if we increase A9X 100% to $75 and discount 50% off Intel Core m3-6y30 to $140, Intel's price is still almost 100% more expensive than A9X.

Source:
How Much Does the Apple Inc. A9X Chip Cost to Make? -- The Motley Fool

http://ark.intel.com/products/family/94027/6th-Generation-Intel-Core-m3-Processors
 
Last edited:
I don't think there is much debate that, at the current snapshot in time, Intel has the process and performance advantage but not the cost/market share advantage. I do think the ARM ecosystem may eventually catch or possibly even surpass Intel in performance as foundry processes catch up, which could happen at 10nm or 7nm.

Talking about the ARM ecosystem. I'm wondering where is the ecosystem among Intel, Microsoft, and those PC manufacturers?

Intel is is competing directly against all PC manufacturers with its own complete server systems, chassis, server motherboards, and RAID products. Microsoft is competing against all PC manufacturers with its own Surface Pro product line.

Intel makes the processor and Microsoft develops the operating system so both of them have technology and cost advantage against all PC manufacturers.

I did a quick check and got the comparison below. The interesting thing is among fabless players, most of them have decent profit. On the Wintel side, that's totally a different story. It looks like Microsoft and Intel are the rich "landlord" and PC manufacturers are the poor "peasant". If this trend continues, Intel and Microsoft probably need to start making a lot more end-user products by themselves. It's because other PC manufacturers can't afford to use Intel and Microsoft products and can't hope to earn meaningful profit by using Wintel solutions.

Yes, transistor density is very important. But if Intel and Microsoft can't find a way to let partners in the Wintel ecosystem to make money, then transistor density will have limited effect in the real market.

[table] style="width: 384px"
|-
| style="height: 20px; width: 106px" | Company
| style="width: 124px" | Profit Margin
| style="width: 155px" | Operating Margin (ttm)
|-
| style="height: 20px" | TSMC
| 36.35
| 38.15
|-
| style="height: 20px" | ARM Holding
| 35.08
| 42.55
|-
| style="height: 20px" | Xilinx
| 24.46
| 30.54
|-
| style="height: 20px" | Apple
| 22.87
| 30.28
|-
| style="height: 20px" | Intel
| 20.63
| 25.93
|-
| style="height: 20px" | Qualcomm
| 20.02
| 28.12
|-
| style="height: 20px" | Avago
| 19.99
| 27.55
|-
| style="height: 20px" | Microsoft
| 8.8
| 13.53
|-
| style="height: 20px" | HP Enterprise
| 4.72
| 7.15
|-
| style="height: 20px" | Asus
| 3.92
| 4.59
|-
| style="height: 20px" | HP Inc.
| 3.71
| 7.82
|-
| style="height: 20px" | Acer
| 0.36
| 0.46
|-
| style="height: 20px" | Lenovo
| -0.29
| 1.18
|-
[/table]



Source: Yahoo Finance.

Note: DELL and GlobalFoundries are private companies. And I need to find a way to analyze Samsung because they make many products among many subsidiaries.
 
Last edited:
My overall point in starting this thread was the Intel can deliver very similar functionality to A9X in a 30% smaller package. That indicates some process advantage. What makes it mind-boggling is how much greater it could be if Intel would optimize the chip layout to fit 24 EUs and dual cores rather than two out of four cores disabled. If there are advantages to the foundry, highly customized model, then Intel has yet to tap those advantages and may become a more formidable competitor if they ever do start to tap them.

hist78: I think there are victims at Asus, Acer and Lenovo, but the villain is not Intel. The victims are the workers who labor under political orders than depress their salaries and seek to employ as many people as possible, producing profit-less, commodity products, in large quantities, so that as many people have as many jobs as possible. The reason I am certain Intel is not the villain is the smartphone business, which is Intel-free, has the same dynamic, with most producers making little or no money, but happily producing jobs for their political masters to crow about.
 
My overall point in starting this thread was the Intel can deliver very similar functionality to A9X in a 30% smaller package. That indicates some process advantage. What makes it mind-boggling is how much greater it could be if Intel would optimize the chip layout to fit 24 EUs and dual cores rather than two out of four cores disabled. If there are advantages to the foundry, highly customized model, then Intel has yet to tap those advantages and may become a more formidable competitor if they ever do start to tap them.

A typical Intel's way (and many other companies too) to maximize its profit is gradually in small increment to deploy or enable improvements throughout product generations and product lines. I'm not surprised that those two cores were disabled. And the reason behind it might not be technical at all. Intel always tries to differentiate each product model and version in order to maximize their profit and grab customers from various markets and various price preference.

Intel can keep doing this way in a free market. But in a free market it will be also a good reason to create new competitors, especially strong competitors. Google, Amazon, Facebook, and Apple don't like their fortune and competitiveness depending on Intel product's roll out plan and incremental improvement.

Think about how many current models and flavors Intel has just for desktop/laptop processor line? There are Core M, i3, i5, i7, i7 Extreme Edition, and vPro and countless speed, number of cores, and cache memory variances.

Yes, Intel definitely can speed up and make more amazing features available sooner. But it will be very likely to reduce Intel's profit significantly. Does Intel's CEO have the courage to do this? Can Intel's cost structure afford to support this change while for so many years Intel has been spoiled by its own fat profit margin?


I think there are victims at Asus, Acer and Lenovo, but the villain is not Intel. The victims are the workers who labor under political orders than depress their salaries and seek to employ as many people as possible, producing profit-less, commodity products, in large quantities, so that as many people have as many jobs as possible. The reason I am certain Intel is not the villain is the smartphone business, which is Intel-free, has the same dynamic, with most producers making little or no money, but happily producing jobs for their political masters to crow about.

My thought is that assume Asus is a bakery using flour provided by Intel and sugar provided by Microsoft. They used to be happy partners doing business in a town called San Francisco. One morning Asus opened its shop for business then found out that Intel started selling cookies using sugar provided by Microsoft in San Francisco and Microsoft started selling cakes using flour provide by Intel in San Francisco too.

Do I expect Asus will happily send "Welcome and Congratulations" cards to Intel and Microsoft?

Back to the real world. Is the unthinkable solution for Intel is to take over all markets and replace all PC manufacturers by itself? Like Apple?
 
Last edited:
hist78 Taking a look at your table of profit margins again, I am thinking about where value comes from. What is the difference between TSMC (at the top) and Lenovo (at the bottom)?

TSMC is creating more value than Lenovo; we know this because Lenovo can be easily replaced while TSMC cannot.
TSMC is in a creative business with IP and innovation. Lenovo is in an assembly business; more about speed, volume and cost of labor.
TSMC is nearer the beginning of the value chain (closer to the innovation), while Lenovo is at the end of that chain (further away).

There are two competitive strategies: Innovation and low cost. Sometimes they are presented as equals, but they are not, at least not in semiconductors. In this industry, innovation is everything; it produces low cost as a byproduct. The better you innovate, the fatter the profit margins get. Your list is a pure scale of innovation.
 
hist78 Taking a look at your table of profit margins again, I am thinking about where value comes from. What is the difference between TSMC (at the top) and Lenovo (at the bottom)?

TSMC is creating more value than Lenovo; we know this because Lenovo can be easily replaced while TSMC cannot.
TSMC is in a creative business with IP and innovation. Lenovo is in an assembly business; more about speed, volume and cost of labor.
TSMC is nearer the beginning of the value chain (closer to the innovation), while Lenovo is at the end of that chain (further away).

There are two competitive strategies: Innovation and low cost. Sometimes they are presented as equals, but they are not, at least not in semiconductors. In this industry, innovation is everything; it produces low cost as a byproduct. The better you innovate, the fatter the profit margins get. Your list is a pure scale of innovation.

By looking the table I created, it's true TSMC makes a lot of money. But along the fabless food chain there are many other companies making decent profit too as shown in the table. Even under tough situation such as Apple does in price and production requirement, there are many Chinese and Taiwanese companies making a lot of money. I called this fabless model a real ecosystem. This is a much healthy environment for most participants of this ecosystem no mater the role is an innovator or not.

On the other hand in the Wintel food chain, other than Microsoft and Intel everyone else is struggling. It's not healthy and can't last forever. Intel and Microsoft might soon find out they are the only two players want to stay in the Wintel game. Unless they make change as soon as possible.

Can Intel or Microsoft make change? Microsoft might be able to do something. But for Intel, IMHO it's very difficult.
 
@benb,
My point is that Intel is not delivering equivalent functionality to the A9X in a 30% smaller package
- Half the A9X die includes silicon functionality that is not likely found in the Core M3-Y
- The Core M3-Y requires another die beyond the one you focused on to achieve full SoC functionality. Granted some of that functionality in the PCH is imposed by Wintel requirements, but nonetheless the full Intel silicon accounting extends well beyond 99mm^2.

But as Don D. noted, it's really difficult to make an unbiased apples to apples comparison.
 
Specification summary:
-Apple A9X vs. Intel Core m3-6y30 processors are roughly comparable
-MHZ: 2260 (Apple), 900-2200 (Intel)
-Die Sq mm: 99 (Intel), 147 (Apple)
-Graphics cores: 24 (Intel), 12 (Apple)
-Graphics performance (PCMark Icestorm Unlimited): 33574 (Apple), 40995 (Intel)
-Cost: Apple iPad Pro ($799), Microsoft Surface 4 Pro ($899)
-Process nodes: Intel 14nm, TSMC "16nm"
Nice try there about cost comparison.
Intel Core m3-6y30 is sold at 281$ (ark).
Apple A9X should definitely cost less than 40$.
It sounds like a 7:1 ratio, at roughly the same performance.


 
hist78: Both ecosystems have their assemblers who exist on thin margins and deep labor cost advantages.
astilo: Two chips: One costs $20 to manufacture (in a 14nm IDM), the other cost $25 (in 16nm foundry) with $250 of fully-loaded costs per unit. Which one is cheaper?
 
Last edited:
astilo: Two chips: One costs $20 to manufacture (in a 14nm IDM), the other cost $25 (in 16nm foundry) with $250 of fully-loaded costs per unit. Which one is cheaper?
Hi benb, I would be really curious to know from where the Intel 20$ figure is coming. And no way the development costs are 10x the manufacturing ones.
 
Back
Top