Intel Broadwell BDW-H delayed May 2015

[Arachnotronic]

Net income Margins of Qualcomm vs Intel :

2013 -

QCOM :
Revenue : 24,866,000
Net Inc : 6,853,000
Net Inc : Revenue = 27.5%

INTC :
Revenue : 52,708,000
Net Inc : 9,620,000
Net Inc : Revenue = 18.3%

Moreover, Qualcomm experienced 25% revenue growth in 2013 vs 2012 while Intel's revenue declined about 1%.

So using your logic, QCOM should go out and start buying Fabs. So should apple, etc.

I don't see that happening. These companies do well by focusing on what they are experts at, not by attempting to own all aspects of production. Inevitably companies that do that become "mediocre" at everything, which is really what I think is happening to Intel. Their tick/tock is producing less and less interesting chips with longer and longer cycle times.

Most of Qualcomm's net income comes from patent licensing/royalties. Operating margins for Qualcomm's QCT are about 20%, well below Intel's PCCG or DCG.

 

[teejee]

.

Fabless model:
Base cost * foundry margin * fabless semico margin = final selling price

"Fab-full" model:
Base cost * margin = final selling price

nope this is wrong. It should be:

Fabless model:
(Base cost foundry * foundry margin) + (base cost fabless * semico margin) = final selling price

"Fab-full" model:
Base cost complete * margin = final selling price

and base cost foundry + base cost fabless is approximately equal to base cost complete. I e the models are the same if foundry, fabless semico and Intel want the same margin.
I e no disadvantage for the fabless model in this regard.
This is basic finance and the whole outsourcing industry in general is based on this.

 

[Homeles]

So using your logic, QCOM should go out and start buying Fabs. So should apple, etc.

I don't see that happening. These companies do well by focusing on what they are experts at, not by attempting to own all aspects of production. Inevitably companies that do that become "mediocre" at everything, which is really what I think is happening to Intel. Their tick/tock is producing less and less interesting chips with longer and longer cycle times.

Ever heard of Samsung? They're doing really well too, and there's a good chance you'll see Qualcomm's share get eaten up by them. Exynos 6 is a custom design, which would remove a lot of Qualcomm's advantage.

Qualcomm definitely has been doing well, but their success is not going to last forever. There's only room for a few players in the market.

Apple's well isolated from the OEM win brawl because they don't sell their SoCs. They'd still benefit from having their own fab. Anand Shimpi's views are similar to my own. He argues that Apple will eventually have to commit to having their own bleeding edge fab. If they want to reduce their reliance on Intel, this would be the way for them to go.

nope this is wrong. It should be:

Fabless model:
(Base cost foundry * foundry margin) + (base cost fabless * semico margin) = final selling price

"Fab-full" model:
Base cost complete * margin = final selling price

and base cost foundry + base cost fabless is approximately equal to base cost complete. I e the models formulas are the same if foundry and fabless semico want the same margin.
I e no disadvantage for the fabless model in this regard.
This is basic finance and the whole outsourcing industry in general is based on this.

I'm sure you're right. Economics is not my strongest subject.

This is one of those subjects that economists fight to death over, though. Vertical integration is the real concept at hand, and it's still a matter of leveraging the strengths it provides. As the number of fabs diminish, the advantages of owning one increases. The industry has consolidated tremendously. IBM's no longer around either.

The two biggest semicos, Intel and Samsung, have been making the changes necessary to leverage their unique positions. Intel's been trying to grab customers for their foundry business, and the success of Altera on Intel's 14nm is critical to attracting more fabless companies. Intel's also done very well to date with exploiting their fab ownership in terms of having great execution. They've all but taken over the x86 market as a result, and they're now applying this to their Atom SoCs.

Samsung's gotten serious about SoC design, and has recognized the importance of custom CPU design to limit Qualcomm's current and past advantage of being the only real SoC supplier with a custom ARM design. This move is also a threat to Apple, as it will make Samsung's flagship designs more appealing. They're allegedly working to reduce their reliance on others for modems. Hopefully for Samsung, they dump can dump Qualcomm in the European and American market. In the meantime, they've started using Intel as an alternative.

Both Intel and Samsung are poised to take some of Qualcomm's market share. Nvidia too, if their Denver plans pay off (if not, they'll be out of the game in no time). I just don't see things ending well for Qualcomm. This is all long term, though.

 

[antihelten]

nope this is wrong. It should be:

Fabless model:
(Base cost foundry * foundry margin) + (base cost fabless * semico margin) = final selling price

"Fab-full" model:
Base cost complete * margin = final selling price

and base cost foundry + base cost fabless is approximately equal to base cost complete. I e the models are the same if foundry, fabless semico and Intel want the same margin.
I e no disadvantage for the fabless model in this regard.
This is basic finance and the whole outsourcing industry in general is based on this.

But the "base cost fabless" equals:

base cost fabless = design cost + base cost foundry * foundry margin

so the final equation is:

(design cost + (base cost foundry * foundry margin)) * semico margin) = final selling price

compared to the "fab-full" model:

(design cost + base cost foundry) * margin = final selling price

So unless the foundry margin for the fabless model is 0%, the fabless model will always have a higher final selling price (assuming similar margins, and costs).

 

[swilli89]

The company that has shipped literally megatons of FinFET silicon.

Wouldn't say anything normally to hyperbole like this but.. you said the word literally. A million tons, huh? So two BILLION pounds of silicon. How many cpu's in a pound I wonder? Man no wonder Intel has that big R&D budget! They have sold more than 1 cup to every single person on this planet. The future is amazing!

 

[Arachnotronic]

Wouldn't say anything normally to hyperbole like this but.. you said the word literally. A million tons, huh? So two BILLION pounds of silicon. How many cpu's in a pound I wonder? Man no wonder Intel has that big R&D budget! They have sold more than 1 cup to every single person on this planet. The future is amazing!

Fixed it, but the point still stands. Intel has shipped hundreds of millions of extremely profitable processors built on the 22nm FinFET process.

How many have the foundry customers or Samsung shipped to end users?

 

[AtenRa]

If an IC costs $1 to produce, the fab will want to sell it with a 30% margin (sometimes more, sometimes less). So the cost to the fabless company in this example is $1.30. They'll want a 30% margin of their own, which is multiplicative, and thus the final price to the OEM is $1.69.

First of all a 30% margin will not be 30 points (Mark up).

If the IC cost is 1 and you want 30% Margin then you will have to sell it at ~$1,42

cost / 100 - (xx% of margin) x 100= price

1 / (100 - 30) x 100 =

(1/ 70) x 100 =

0,0142 x 100 =

$1,42

You can confirm this by deducting 30% from the 1,42 that should get you close to 1.

Intel on the other hand likes their 60% margins. Their IC costs $1.00 to make as well, but they can sell their IC for their desired $1.60 and still come ahead.

With 60% margin you will sell it at $2.5

Fabless model:
Base cost * foundry margin * fabless semico margin = final selling price

"Fab-full" model:
Base cost * margin = final selling price

You forgetting the R&D investment to create the Fab and the Process and you also dont count the wafer volume production per month.

Lets say you spend the same R&D for Intel and TSMC for the next process, but one of the two has higher volume production than the other. That means this company can sell lower from the start or have a ROI faster.

Also, Intels 60% margins include both the production and the IC design. TSMC has lower margins but they only manufacture the IC, they dont design it. NVIDIA designs the IC and is the final seller of the product, Intel is both the designer and manufacturer.

 

[NTMBK]

Fabless model:
Base cost * foundry margin * fabless semico margin = final selling price

"Fab-full" model:
Base cost * margin = final selling price

Remember though, that the fab-owning margin will need to be larger than either of the individual foundry and fabless margins. That margin has to fund R&D and capital expenditure for both the next generation of fabs and the next generation of processor design.

 

[teejee]

But the "base cost fabless" equals:

base cost fabless = design cost + base cost foundry * foundry margin

so the final equation is:

(design cost + (base cost foundry * foundry margin)) * semico margin) = final selling price

compared to the "fab-full" model:

(design cost + base cost foundry) * margin = final selling price

So unless the foundry margin for the fabless model is 0%, the fabless model will always have a higher final selling price (assuming similar margins, and costs).

this is simply wrong, please think about this once more. outsourcing does NOT mean that your margin measured in % decreases. It only means that the size of your business meaured in dollar decreases.

 

[Idontcare]

this is simply wrong, please think about this once more. outsourcing does NOT mean that your margin measured in % decreases. It only means that the size of your business meaured in dollar decreases.

I suspect that you are going to need to go beyond simple implied outcomes and actually work out an example of die prices for AMD versus Intel before folks can take this conversation beyond the theoretical and into the applied.

This may work against your argument, or for it, but the next step here is self-evident and likely required (albeit perhaps academic).

 

[mikk]

So you mean that since Broadwell-H/K will be released in summer of 2015, it gives us desktop Skylake two years after that, i.e. in summer of 2017?

First LGA desktop Skylake models are planned for Q2 2015 according to newest infos. It means desktop users can choose between Broadwell-K (GT3 graphics+edram, free multiplicator and Skylake (GT2, closed multiplicator) in summer next year.

According to Intel's desktop roadmap, Skylake desktop Core i5 and i7 microprocessors, using Skylake-S core, along with 100 series chipset will be launched in the second quarter 2015. Broadwell desktop i5/i7 CPUs will be also released in Q2 2015. What is interesting is that all unlocked processors at launch will be Broadwell based, and all locked parts will be built on Skylake-S core. At this time, Intel does not plan to transition Celeron, Pentium and Core i3 processors to Skylake architecture before second half of 2015.

http://www.cpu-world.com/news_2014/2014052601_Intel_Skylake_desktop_CPUs_to_launch_in_Q2_2015.html

 

[frozentundra123456]

First LGA desktop Skylake models are planned for Q2 2015 according to newest infos. It means desktop users can choose between Broadwell-K (GT3 graphics+edram, free multiplicator and Skylake (GT2, closed multiplicator) in summer next year.


http://www.cpu-world.com/news_2014/2014052601_Intel_Skylake_desktop_CPUs_to_launch_in_Q2_2015.html

That sounds like bad news to me if true. Many were hoping for a decent performance improvement from skylake, but if the unlocked chips are broad well, it sounds like either skylake in not much faster, or they are allowing the unlocked desktop models from each generation to be released later like they are doing now with the HEDT platform.

 

[antihelten]

this is simply wrong, please think about this once more. outsourcing does NOT mean that your margin measured in % decreases. It only means that the size of your business meaured in dollar decreases.

Outsourcing means you pay someone else to do something for you instead of doing it yourself, those someone else expect to have a positive margin for themselves (or they wouldn't be in business). The ones paying that margin is you, hence an increased total cost compared to doing it yourself, all else being equal.

Now the reason companies still outsource is because they generally can't do the thing at the same cost (normally due to lack of volume, when talking about high-tech stuf like fab business). And as such the total cost even with the extra margin on top is still lower than doing it yourself.

As far as the topic goes however, Intel arguably can do it ("it" being the fab side of things) at a cost comparable to TSMCs.

edit: although I agree that if you consider say Qualcomm and TSMC as a single entity then you are right that the margin stays the same, and one could argue that in the competition against Intel, they are practically speaking a single entity since they are dependent upon each other.

 

[Sweepr]

First LGA desktop Skylake models are planned for Q2 2015 according to newest infos. It means desktop users can choose between Broadwell-K (GT3 graphics+edram, free multiplicator and Skylake (GT2, closed multiplicator) in summer next year.


http://www.cpu-world.com/news_2014/2014052601_Intel_Skylake_desktop_CPUs_to_launch_in_Q2_2015.html

Seems like I was right and people expecting Broadwell-K to delay the whole dektop Skylake line-up to 2017 were way off.

That's an interesting piece of news though, it means Skylake is probably coming to desktops first, while mobile parts should launch sometime in H2.

 

[ShintaiDK]

Seems like I was right and people expecting Broadwell-K to delay the whole dektop Skylake line-up to 2017 were way off.

That's an interesting piece of news though, it means Skylake is probably coming to desktops first, while mobile parts should launch sometime in H2.

Skylake will increase the pwoer consumption slightly compared to Broadwell. So its not a surprise. Well, it is for "some".

So Q2 2015 for DDR4, PCIe 4.0 and 512bit AVX 3.2.

 

[mikk]

Seems like I was right and people expecting Broadwell-K to delay the whole dektop Skylake line-up to 2017 were way off.

That's an interesting piece of news though, it means Skylake is probably coming to desktops first, while mobile parts should launch sometime in H2.

Yes I expected the same for desktop. I didn't expect it so close with Broadwell-K though.

Skylake will increase the pwoer consumption slightly compared to Broadwell. So its not a surprise. Well, it is for "some".

So Q2 2015 for DDR4, PCIe 4.0 and 512bit AVX 3.2.

For consumer Skylake the last we heard there is no AXV3 and no PCIe 4.0 as well. Not a big deal because AVX2 isn't used in consumer applications apart from stress testers like Prime and maybe some synthetic benchmarks.

 

[jpiniero]

Yeah, I am not buying it. So they are going to release Skylake on desktops 6 months before any other models and only the locked desktop i5 and i7? That's like maybe 5% of Intel's mainstream sales.

 

[NTMBK]

For consumer Skylake the last we heard there is no AXV3 and no PCIe 4.0 as well. Not a big deal because AVX2 isn't used in consumer applications apart from stress testers like Prime and maybe some synthetic benchmarks.

No AVX3? Could you link me to that, please? That would be a very big deal because it would imply a different core design for the server and consumer parts.

 

[mikk]

No AVX3? Could you link me to that, please? That would be a very big deal because it would imply a different core design for the server and consumer parts.

Intel AVX-512 in Intel products
Intel AVX-512 will be first implemented in the future Intel® Xeon Phi™ processor and coprocessor known by the code name Knights Landing, and will also be supported by some future Xeon processors scheduled to be introduced after Knights Landing.

https://software.intel.com/en-us/blogs/2013/avx-512-instructions


It could be that only highend Server Skylake gets it.

 

[ShintaiDK]

You cant exclude desktops just because it says Xeons where the context is purely targetted at server segment.

 

[NTMBK]

https://software.intel.com/en-us/blogs/2013/avx-512-instructions


It could be that only highend Server Skylake gets it.

That quote simply says that Knight's Landing will come first, not that consumer Skylake will not support it. And don't forget, the Xeon E3 series is just rebadged consumer APUs.

 

[mikk]

That quote simply says that Knight's Landing will come first, not that consumer Skylake will not support it. And don't forget, the Xeon E3 series is just rebadged consumer APUs.

It says Knights will support it and some future Xeon processors. Because it says scheduled after Knights landing consumer Skylake is most likely not the aim.

 

[Homeles]

It says Knights will support it and some future Xeon processors. Because it says scheduled after Knights landing consumer Skylake is most likely not the aim.

Seems silly to not have it enabled for consumers. AVX-512 would take up a very significant portion of the core.

 

[witeken]

Seems like I was right and people expecting Broadwell-K to delay the whole dektop Skylake line-up to 2017 were way off.

That's an interesting piece of news though, it means Skylake is probably coming to desktops first, while mobile parts should launch sometime in H2.

Very interesting. Q2 is even earlier than what BK told us in the Q1 conference call. This probably also means 10nm in Q2'16.

 

[mikk]

It means nothing for 10nm.