Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intels-mind-boggling-process-advantage-explained.7871/page-2
        )

    [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

There's nothing "mind boggling" about it. Intel are willing to push the process to the limit and use very tight and restrictive design rules, which impact yield and design difficulty, in order to get the smallest area highest performing CPUs on the most expensive wafers with reduced yield, because this is what makes sense for their high-margin locked-into-x86 only-a-few-designs business. TSMC use more relaxed rules and process to get better yield, easier design, and cheaper wafers, because this is what makes sense for their lots-of-designs lower-cost foundry business.

So Intel have a performance advantage and TSMC have a price advantage, because their targets are different. What's mind-boggling about that?
 
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?


First, I didn't include Foxconn (Hon Hai Precision Industry Co., Ltd) in my table. Foxconn anlog with several other companies (such as Quanta Computer, Winstron, Compal, Pegatron, and many other) assemble desktops and laptops on behalf of Apple, HP, Lenovo, Asus, and Acer. Foxconn, Quanta, Winstron, Cpmpal, and Pegatron are the "assembler" you're talking about and they do have thing profit margin.

Those assemblers don't make meaningful profit while HP, Asus, Acer, Lenovo, and very possible DELL don't make meaningful profit either. So who are making big money in this Wintel ecosystem?

Here is one of the biggest problem in the Wintel ecosystem: only Intel and Microsoft are enjoying the wonderful profit and they don't want to give up an inch to other players in the Wintel ecosystem. In some cases Intel and Microsoft even compete directly against HP, Lenove, Asus, Acer, and Dell who are using Intel and Microsoft's solutions. Intel's server products and Microsoft's Surface Pro are the bloody evidences.

This Wintel ecosystem is more and more like a "peasant" system.

Then we can look into the fabless ecosystem. Between the TSMC who makes the processors/SoC and the Foxconn who assemble iPhones, there are many other companies making decent profit. This is a much healthy ecosystem and it will develop many more profitable and sizable markets.

The $20 Intel CPU cost you mentioned is great, but only good for Intel. For Acer, Asus, Lenove, HP, and DELL, they won't see that $20 unit price when they receive invoices from Intel. The real price they got from Intel is probably five to ten time of that $20.

It makes me wonder who doesn't want to do away from Intel's products if building your own processors can differentiate youself from your peer? Especially if you have more money in the bank than Intel or you are much bigger than Intel!

Apple can pay less than $40 to get the A9X instead of paying $100 ~ $200 for a Intel Core m3-6y30. Apple is again and again telling the world the huge benefit of design your own processors/SoC.

Market capitalization:

Apple $528.19B
Google $483.5B
Microsoft $393.26B
Facebook $331.71B
Amazon $328.75B
Intel $142.75B
TSMC $125.8B
AMD $3.21B

Source: Yahoo Finance. Data collected on May 25, 2016.
 
Last edited:
There's nothing "mind boggling" about it. Intel are willing to push the process to the limit and use very tight and restrictive design rules, which impact yield and design difficulty, in order to get the smallest area highest performing CPUs on the most expensive wafers with reduced yield, because this is what makes sense for their high-margin locked-into-x86 only-a-few-designs business. TSMC use more relaxed rules and process to get better yield, easier design, and cheaper wafers, because this is what makes sense for their lots-of-designs lower-cost foundry business.

So Intel have a performance advantage and TSMC have a price advantage, because their targets are different. What's mind-boggling about that?

You forgot about SPARC M7 in your theory. Processor manufactured using TSMC 20nm, that is few times more powerful than any existing 22/14nm Intel Xeon today.

Or 16nm Pascal GP100 vs 14nm Intel KNL (5,3 vs ~3 TFlops).

So Intel do not have any performance advantage. Their only advantage is that they are "king of the hill" in x86 market.
 
Is Intel impoverishing their customers? Probably more than TSMC is. The foundry model is more promiscuous--you can have more than one partner. It keeps greed in check. Whereas with Intel, they have pricing power and little fear of substitution. So they can be more greedy.

I acknowledge this, in part, because it's a good thing. TSMC also has pricing power, and it's also a good thing. The dynamics are little different because the pricing power is more limited by the risk of substitution, which puts a limit on greed, but both Intel and TSMC deserve what they get. It isn't unearned or unfair.

What is driving the low costs in the foundry world is sharing IP. That is also a good thing; it's efficient. I think there is some forcible sharing though, semi-stealing, that has pushed Qualcomm into an arrangement where they make their advanced chips mainly in the US, joining Apple. It is why Apple doesn't commit fully to TSMC, I think. Sometimes this semi-stealing is justified with the false "peasants system" logic. Of course, this logic is circuitous and if enacted, would be self-defeating also.
 
Is Intel impoverishing their customers? Probably more than TSMC is. The foundry model is more promiscuous--you can have more than one partner. It keeps greed in check. Whereas with Intel, they have pricing power and little fear of substitution. So they can be more greedy.

I acknowledge this, in part, because it's a good thing. TSMC also has pricing power, and it's also a good thing. The dynamics are little different because the pricing power is more limited by the risk of substitution, which puts a limit on greed, but both Intel and TSMC deserve what they get. It isn't unearned or unfair.

What is driving the low costs in the foundry world is sharing IP. That is also a good thing; it's efficient. I think there is some forcible sharing though, semi-stealing, that has pushed Qualcomm into an arrangement where they make their advanced chips mainly in the US, joining Apple. It is why Apple doesn't commit fully to TSMC, I think. Sometimes this semi-stealing is justified with the false "peasants system" logic. Of course, this logic is circuitous and if enacted, would be self-defeating also.
.

I don't have problem with how Intel, Microsoft, or TSMC set their price. It's a free market anyway. But I do have problem with Intel and Microsoft compete directly against their very own customers in the end user market, such as Intel's servers and Microsoft's Surface Pro.

Can you imagine if Qualcomm starts selling Qualcomm's own smartphone? Or TSMC starts selling a processor design by TSMC's R&D staff.
 
You forgot about SPARC M7 in your theory. Processor manufactured using TSMC 20nm, that is few times more powerful than any existing 22/14nm Intel Xeon today.

Or 16nm Pascal GP100 vs 14nm Intel KNL (5,3 vs ~3 TFlops).

So Intel do not have any performance advantage. Their only advantage is that they are "king of the hill" in x86 market.

You're confusing chip performance (architecture, circuits etc.) with process performance (transistors, interconnect etc.). The headline was "Intel's mind-boggling PROCESS advantage explained". My point is that the target market for which the process is optimised is different for Intel (x86 CPUs) and TSMC (everything else), not that Intel's CPUs are better than any other CPU ever produced for every application.
 
Last edited:
This thing about competing with customers is valid, but ethics doesn't cover it. There is no Ten Commandments of Capitalism. It's simply a judgement call. By competing with your customers, you lose some customers. If you can make more money, by taking this route, then it makes sense.

It's somewhat sketchy to make this argument about purity and not competing with customers though, since Microsoft has been in hardware for decades. They made shitty hardware for decades and lost money (Lumia, Zune). But with some leasons learned, they finally get it right, invent a new category of device that even Apple is copying, and suddenly it's impure of them and how dare they?

For innovation to meaningful, it must break some eggs and wipe out the old way of doing things. I don't buy into the idea that Microsoft or any company is obligated to bring old, ailing customers along into the new world, out of the goodness in their hearts.
 
Last edited:
IanD: Trust me, i really wanted to do it this way. But I do not have any real unbiased parameters.

But i remember one quote, which i like btw.

You really need to take a look at the performance of the devices and what's being delivered.

Brian Krzanich, Intel CEO

So i compared similar product aiming similar market designed to perform similar tasks.

Sorry if you don't like it or if you think, that this is not accurate, but i believe, that it is still more accurate than comparison of passive cooled A9(X) and active cooled Core M.

Or comparing processes according to gate-metal pitch, or even worse, based on Intel's presentation slides.

ps.: Architecture? Architecture of Intel's processors is considered the best on the market. And circuits (layout)? Intel is IDM so who is responsible for this fault?
 
This thing about competing with customers is valid, but ethics doesn't cover it. There is no Ten Commandments of Capitalism. It's simply a judgement call. By competing with your customers, you lose some customers. If you can make more money, by taking this route, then it makes sense.

It's somewhat sketchy to make this argument about purity and not competing with customers though, since Microsoft has been in hardware for decades. They made shitty hardware for decades and lost money (Lumia, Zune). But with some leasons learned, they finally get it right, invent a new category of device that even Apple is copying, and suddenly it's impure of them and how dare they?

For innovation to meaningful, it must break some eggs and wipe out the old way of doing things. I don't buy into the idea that Microsoft or any company is obligated to bring old, ailing customers along into the new world, out of the goodness in their hearts.

I agree with you for the most part. Like I said before, it's a free market anyway. The problem I see is that in business the relationship is two-way street. If Intel keeps garbing their customers' nice lunches and at the same time asking customers to support Intel's fat profit by paying premium price, I don't see it's a healthy ecosystem. It's just a "peasant" system.

We can say Intel's growing profit and revenue from server division is the result of the fast growing Internet, clouding computing, and AI. But I have to say a big portion of that is because Intel ran out options about how to grow their business and server products is the one they can still increase revenue significantly (even though several their important customers need that too).

It's fine if Intel insists to stay on this business model but they will be forced to become a system company like the old IBM. It's their choice.

A real ecosystem is not a marketing gimmick. A real ecosystem is a long term commitment and mutually beneficial relationship. By looking into a section of Q&A of 3Q2015 Xilinx conference call, I hope people can sense what kind of the relationship between TSMC and Xilinx is. Can we expect DELL, Lenovo, HP, Asus, or Acer to say something like this to support Intel when their own products are in trouble?


Ambrish Srivastava (Analyst - BMO Capital Markets):Hi, thank you. Moshe, I just had a clarification on the FinFET comment you made. Is this an issue emanating from the difficulty, in your opinion, TSMC is having or is it coming from the Xilinx side? A just a quick follow-up -- what is the timing from tapeout to when we should expect production volume? Thank you.

Moshe Gavrielov (CEO):
There are no issues with TSMC. They have had numerous tapeouts already. They're giving us full support. The design, whenever you encounter a new generation of product, tends to unearth problems that you did not anticipate, and, as a result, the closing of all of these issues is taking a little longer, plus the challenges related to designing for FinFET transistors are more significant. So, it is not a TSMC challenge or issue at all. It's just our ability to finish the design with their support.

Source: Xilinx (XLNX) Earnings Report: Q3 215 Conference Call Transcript - Pg.6 - TheStreet
 
Last edited:
It seems to me having read this thread several times that Intel's mind boggling advantage is nothing more than the scaling advantage of Intel's 14nm process over TSM's 16nm, which Is well documented elsewhere on Semiwiki. What is much more mind boggling to me is their utter failure to translate this advantage into mobile product, where not only has Krzanich spent over $10 billion in support of his undertaking to the Board on promotion to CEO but also has converted what used to be a "three year" lead into one that is little more than nine months. More to the point Murthy cancelled Broxton because it was at least two years behind. How it can be deduced that Apple would get an advantage from using a SoC that imtel
had never been able to make is quite beyond my powers of logic. In a few months time TSMC will be manufacturing the A10x, on which Mark Hibben supported by DAN has suggested that Apple has invested capex of $2 billion.
 
So to summarize what the forum thinks on this topic:
1) No one seems to directly dispute the 30% or so scaling advantage Intel is holding on to, but some say it's irrelevent, since a 30% scaling advantage doesn't result in a 30% lower fully loaded cost meaning the scaling advantage can be neutralized by lower costs elsewhere in the cost stack.
2) I believe the source of the lower overall cost in the foundry world is won through the sharing of ARM-based IP, which greatly lowers the costs for low and mid-range chips. While this thread compared A9X with Core m3, the A9X is a riff on the excellent A57 and A72 cores, which have lower performance, but not dramatically lower. Tweaking a good design is easier, faster and less costly than go-it-alone like Intel does.
3) TSMC partisans pointed out SPARC and Pascal GPUs as examples of TSMC high-performance parts, but the thread doesn't investigate performance, since that is more of an output than an input. What is more interesting is the details of the inputs required to produce constant performance, as we approximately have with A9X and Core m3.
4) TSMC partisans incorrectly pointed out that Core m3 requires active cooling. It is a common mistake though because most Core parts require cooling fans. Except for Broadwell-Y and now Skylake-Y which are special fanless tablet/notebook chips. Core m3 is a Skylake-Y part.
5) There was discussion about the nature of competition in the IDM world vs. in the foundry world. The foundry world seems to have fewer "peasants". The fairness of profitability of Intel and Microsoft compared to Intel and Microsoft dependents was questioned.
6) As evidence of the fully loaded cost of A9X I used the sale price of the iPad Pro 12" ($799). As evidence of the fully loaded cost of Core m3, I used the Surface Pro 4 with Core m3 ($899). These were within $100 of each other suggesting Intel's scaling advantage has in fact been neutralized, and maybe more than neutralized. It could indicate different margins and specifications though. It is a very rough comparison, used only because there are no A9X list prices, and Core m list prices are the same for m3, m5 and m7, suggesting it is meaningless.
 
Last edited:
@benb,
I disputed your contention of a 30% scaling advantage, but you found it convenient to neglect. Key points:
- 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.

Furthermore, if you look at Intel's own public calculations, you'll find that;
- TSMC and Samsung A8 and A9 processors have greater raw transistor densities for "comparable parts"
- The disposition of transistors (how they are used) and transistor sizes are different between the Apple parts and Intel parts - one is optimized for speed, the other for density/integration.
 
I ask you simple question: Do you have this fanless Surface 4?

Because if yes, then you should know that it is still throttling so badly, and at clock around or under 1GHz is far behind A9X which holds 2+GHz.

It have to run at least at 2GHz to be competitive but it is not possible without active cooler. Just look at TDP tables. 3,5W for 600MHz, 4,5W for 900 and 6W for 1100MHz... And note, that in Intel's terminology TDP means typical, not maximum TDP.

And why are you still ignoring, that A9X has much more integrated peripherals than Core's, when you are talking about size of die?

btw.: don't mark any person from south-eastern EU as partisan, in any context. Just friendly advice.
 
In the end for CPUs it all comes down to power and cost per MIP/GOP (or whatever metric you want to pick to measure "value"). Right now at the high-end (server/laptop) Intel with x86 is the leader, but at the medium/low-end Apple/Qualcomm/Huawei...(pick your favourite smartphone) with ARM are winning, because the architecture and wide variety of CPU cores give a much more scaleable solution than x86.

It's very difficult to see how anything can knock the ARM architecture off its perch for medium/low end applications, but the particular winner in this space can easily shift from one year to the next. It's easy to see how -- at least in theory -- Intel could run into trouble in the high-end market, if either AMD/Zen or one of the multicore ARM vendors manages to provide a solution which is lower power or cheaper than Intel for the same performance, and this is made worse for them if TSMC manage to deliver their 7nm process on time because this is very similar to Intel's proposed 10nm process (yes, I know neither is really 7nm or 10nm, they're just marketing names).

Intel giving up on mobile after spending tens of billions trying to get in means the're perilously close to becoming a one-trick pony, most of their profits now come from the server division and this is also at risk if they can't keep ahead -- which is less likely when they've lost their process advantage. There's definitely some risk that they'll be the next Nokia...
 
There's another way to look at this besides "tisk tisk how the mighty have fallen". One way is the Christensen approach, which is definitely a "tisk tisk" narrative; lower cost/lower performance parts supplant high cost/high performance parts by creating new markets. That is definitely what happened in mobile. The next phase of that narrative is beginning to take shape: The "march north" of the low cost/low performance part into higher and higher performance niches, using the cost advantage to breach the market stronghold and destroy the incumbent.

However, history rhymes but rarely follows a script. My thinking is, the cost advantage in the foundry world is found in Cortex A parts, not Kyro or Twister or Tegra X1. Cortex A has hundreds of licensees, whereas, the others are customized for a single customer. Being customized, and unique, means the costs are concentrated. Products based on Twister, Kyro, Core or Tegra X1 will seek similar margins, performance, sockets and price points. I'm not saying identical in any of those parameters and there will be advocates for one vendor or another, foundry vs. IDM, but my point is, it's pretty close. My point is, I distinguish the custom parts business, which requires high margins due to high costs, from the Cortex A foundry business. And I think Intel is not particularly disadvantaged when facing custom part competition.

If Cortex A parts move into server performance profiles, and bring hundreds of licensees, then we'll see the Christensen process play out as scripted, as the economics of shared costs stays in place. If, however, it proves difficult to interest hundreds of licensees in winning a piece of a mature market, the cost advantage will vanish and it'll be a stalemate with the winner most likely being the first mover--in this case, Intel.
 
Last edited:
Back
Top