The Intel Atom Thread

[liahos1]

TSMC has about 100k 28nm wafer capacity per month. or roughly 20% of their overall capacity, 30% of their q2 revs. This comes to about $4.4k per wafer in revenue. Their corp average gross margin 45% or rougly $2.4k in all in wafer costs.

527 die (90% yield on a 102mm^2 die) on 2.4k of wafer costs = 4.5 cost per die. What i forgot to do was count packaging costs. Assuming packaging is 30% of of the overall COGS

$4.5/.7 = $6.5. So yes you can sell these for $8-10 dollars profitably. the marginal costs of producing this is going to be very very low given a large percentage of the 4.5 cost / die is non cash. When Merrifield comes out it will likely be smaller or equivalent to s600 in size (but better in perf/watt). If intel wants a price war it can most assuredly win one. I think ultimately what intel needs to succeed in mobile isnt a better SOC as all the device managers I talked to at the various OEMS were very pleased with Intel's SOC's. Its more of the handholding that intel is still coming up to speed with, with regard to OEMS/Carriers. That is not a structural issue though. Overtime intel will learn this market and drive out all the weak hands.

 

[Abwx]

You really think Intel is selling these chips at negative gross margin?

Unlikely.

My point is that selling them under 20$/chip , that is
about 25$ once packaged, will hardly pay for the investments ,
the only advantage of very low prices and 22nm for thoses
chips is to eventually keep the capacity productions means
at reasonable output if ever their traditionnal markets
keep shrinking or stagnating.

Usualy Intel kept their factories running at full capacity
thanks to constant die sizes through node shrinks but
with the current trend their new CPUs are slightly smaller ,
hence they have excess capacity that is dumped on whatever
product that can sell within productions costs.

 

[Abwx]

TSMC has about 100k 28nm wafer capacity per month. or roughly 20% of their overall capacity, 30% of their q2 revs. This comes to about $4.4k per wafer in revenue. Their corp average gross margin 45% or rougly $2.4k in all in wafer costs.

527 die (90% yield on a 102mm^2 die) on 2.4k of wafer costs = 4.5 cost per die. What i forgot to do was count packaging costs. Assuming packaging is 30% of of the overall COGS

$4.5/.7 = $6.5. So yes you can sell these for $8-10 dollars profitably. the marginal costs of producing this is going to be very very low given a large percentage of the 4.5 cost / die is non cash. When Merrifield comes out it will likely be smaller or equivalent to s600 in size (but better in perf/watt). If intel wants a price war it can most assuredly win one. I think ultimately what intel needs to succeed in mobile isnt a better SOC as all the device managers I talked to at the various OEMS were very pleased with Intel's SOC's. Its more of the handholding that intel is still coming up to speed with, with regard to OEMS/Carriers. That is not a structural issue though. Overtime intel will learn this market and drive out all the weak hands.

You seem unaware that Intel process is much more expensive
than TSMCs....

I give you a first hand hint for your future estimations :

During Q1 2014 production cost of a 22nm Finfet chip with 100mm2
area will shrink to 20$ , cost of a same process based 200mm2 chip ,
about a Haswell , will be 50$ at the same date.

 

[mikk]

The Iris 5100 that goes into the 28W SKUs seems to be quite competitive with Trinity's top Radeon HD 7660G in actual game benchmarks. And peak power consumption appears similar for both in the 40-45W range.

Iris 5100 - http://www.notebookcheck.net/Intel-Iris-Graphics-5100.91977.0.html
Radeon HD 7660G - http://www.notebookcheck.net/AMD-Radeon-HD-7660G.69830.0.html

The main point is, AMD only ships one model in full glory with all shaders. A8, A6, A4 models are crippled unlike Intels GT2. Even the top notch Trinity is just equal to Haswell GT2. To say Trinity is faster is just plain wrong. On the desktop side it's true but in the much more important mobile space it is plain wrong. Furthermore AMD notebooks are often coupled with a second GPU which is crap because of heavy micro stuttering.

 

[mikk]

You seem unaware that Intel process is much more expensive
than TSMCs....

Can you prove it this time?

 

[Abwx]

Can you prove it this time?

I already posted theses numbers and quite a lot of people
think that it was pulled out of thin air but i wouldnt dare
to post random numbers.

Theses are extracted from a study of different process
costs published by the FDSOI consortium , the paper
can be found on the net.

The same paper stipulate that Finfets will manage
to have lower cost than bulk HKMG if ever the cycle
is longer than two years , hence Intel s is likely
to extend it to three years to benefit from said
cost advantage past H1 2014.

 

[Khato]

Theses are extracted from a study of different process
costs published by the FDSOI consortium , the paper
can be found on the net.

Sounds like an extremely reliable source for such detailed information on Intel's process technology... Need I be more sarcastic? Or is that sufficient?

 

[liahos1]

I already posted theses numbers and quite a lot of people
think that it was pulled out of thin air but i wouldnt dare
to post random numbers.

Theses are extracted from a study of different process
costs published by the FDSOI consortium , the paper
can be found on the net.

The same paper stipulate that Finfets will manage
to have lower cost than bulk HKMG if ever the cycle
is longer than two years , hence Intel s is likely
to extend it to three years to benefit from said
cost advantage past H1 2014.

amazing you are quoting analysis by the FDSOI consortium, the same consortium that is now scramblign to catch up to intels 4 year lead in commercialized finfets. Intel's wafer costs are lower than TSMC's. I'm at home eating dinner now and dont want to waste more time trying to convince you otherwise. Good luck!

 

[Abwx]

Sounds like an extremely reliable source for such detailed information on Intel's process technology... Need I be more sarcastic? Or is that sufficient?

Sarcasm is all that is left when there is no argument
and not even the clue that every foundry get its tools
at the same place as Intel , namely at ASML mainly ,
they have really no problem to estimate the costs
since it s the core of their very business.

amazing you are quoting analysis by the FDSOI consortium, the same consortium that is now scramblign to catch up to intels 4 year lead in commercialized finfets. Intel's wafer costs are lower than TSMC's. I'm at home eating dinner now and dont want to waste more time trying to convince you otherwise. Good luck!

Intel s lead has more to do with money that with know how.

The smaller the process the more lithography and patterning
tools to keep the manufacturing flow continous.

Not every foundry has the mean to double or triple
the waffers patterning for instance.

Edit : so much for Intel cost being less than TSMCs ,
i guess that you have no clue of how a chip
is manufactured.

 

[SiliconWars]

We keep hearing about this process lead (which seems to increase with every post - 4 years now really?) that Intel has, yet the actual chips seem to be missing...

I would love to know what Intels 22nm and finfets are supposed to be doing for them if the chips are the same size, performance and power as the competitors on their "inferior" 4-year-older process. There must be a simple explanation that is evading me.

 

[mikk]

I already posted theses numbers and quite a lot of people
think that it was pulled out of thin air but i wouldnt dare
to post random numbers.

No problem I didn't expect something reliable from you.

 

[Sweepr]

The Iris 5100 that goes into the 28W SKUs seems to be quite competitive with Trinity's top Radeon HD 7660G in actual game benchmarks. And peak power consumption appears similar for both in the 40-45W range.

Iris 5100 - http://www.notebookcheck.net/Intel-Iris-Graphics-5100.91977.0.html
Radeon HD 7660G - http://www.notebookcheck.net/AMD-Radeon-HD-7660G.69830.0.html

Yup, and lets not forget that 28W Haswell ULT completely destroys any 35W AMD mobile offering when it comes to CPU performance, just a quick CineBench 11.5 comparison:

A10-5750M: 2.32 Multi / 0.85 Single.
Core i7 4558U: 3.48 Multi / 1.45 Single.

 

[Phynaz]

You seem unaware that Intel process is much more expensive
than TSMCs....

I give you a first hand hint for your future estimations :

During Q1 2014 production cost of a 22nm Finfet chip with 100mm2
area will shrink to 20$ , cost of a same process based 200mm2 chip ,
about a Haswell , will be 50$ at the same date.

You keep pulling Intel production costs out of thin air without providing any proof. We aren't falling for it.

 

[Abwx]

We keep hearing about this process lead (which seems to increase with every post - 4 years now really?) that Intel has, yet the actual chips seem to be missing...

I would love to know what Intels 22nm and finfets are supposed to be doing for them if the chips are the same size, performance and power as the competitors on their "inferior" 4-year-older process. There must be a simple explanation that is evading me.

Actualy a lot of people are arguing out of complete ignorance
of how the industry work let alone of how a chip is manufactured ,
as for talking of what interest us to try to grasp the process
advantages and drawbacks it is litteraly science fiction that will
be answered at the same level , that is either ad hominem or more
simply plain denial out of the unability to only understand what
is discussed , never mind , it doesnt glorify Intel so it can only
be viral marketing or at the very least deffamation....

 

[Abwx]

You keep pulling Intel production costs out of thin air without providing any proof. We aren't falling for it.

Dare to give your own estimations.??..

If i m wrong tell us by how much.

 

[blackened23]

Actualy a lot of people are arguing out of complete ignorance
of how the industry work let alone of how a chip is manufactured ,
as for talking of what interest us to try to grasp the process
advantages and drawbacks it is litteraly science fiction that will
be answered at the same level , that is either ad hominem or more
simply plain denial out of the unability to only understand what
is discussed , never mind , it doesnt glorify Intel so it can only
be viral marketing or at the very least deffamation....

LOL, tell us more about this "viral marketing" that you speak of.

 

[Khato]

Sarcasm is all that is left when there is no argument
and not even the clue that every foundry get its tools
at the same place as Intel , namely at ASML mainly ,
they have really no problem to estimate the costs
since it s the core of their very business.

Not really - I know full well what they can base their estimates on (and even those are estimates as last I checked they don't have firm figures for Intel's equipment purchases. Then there's the fact that they happily skew those estimates as far as they believe they can get away with while still appearing realistic in order to make their technology look superior. Hence why sarcasm is indeed quite warranted.

 

[Enigmoid]

Yeap, two years later and Intel still loose to Trinity's iGPU in the same price point. What a turd

HD 4600 and llano are quite close and back when llano debuted you would never have expected intel to make such a massive jump. If we see another HD 3000 -4000 like jump on the 14 nm node shrink trinity/richland level performance is certainty possible.

It must get you to the nerves that a two year old Trinity is still faster than Haswell HD4600. :whiste:

Trinity is not two years old. Mobile was released well before desktop which was released just over a year ago.

Care to show us some numbers at the same TDP ?? say 35W ???

http://www.anandtech.com/show/7106/...t-1-the-apu-and-radeon-hd-8650g-performance/3

The a10-5750m is significantly faster than the 15w haswell i7 at playable settings. Considering how tdp limited haswell ULT is, at 35 watts haswell is going to put up quite a fight. The 28 watt iris 5100 model is probably quite a bit better better.

And its no one but AMD's fault that they don't sell 45w cpus. No direction in this company. You would think they would sell mobile CPUs with a higher tdp considering how tdp limited their designs are and how easily a larger notebook can cool a 45 watt cpu.

 

[liahos1]

LOL, tell us more about this "viral marketing" that you speak of.

lol this is too much. Abwx tells us more tales. Maybe the tale where amd magically usurps ipc leadership or where tsm actually gives us a real shrink btwn 2014-2017. Tell us about how magical fdsoi is and how the common platform is the best

 

[Abwx]

Not really - I know full well what they can base their estimates on (and even those are estimates as last I checked they don't have firm figures for Intel's equipment purchases. Then there's the fact that they happily skew those estimates as far as they believe they can get away with while still appearing realistic in order to make their technology look superior. Hence why sarcasm is indeed quite warranted.

You re looking too much at the PR side of thing and i m not interested
by marketing related material , actualy you re understimating thousands
of scientists and engineers that have full understanding of processes
and subsequent devices caracteristics as well as manufacturing tools
requirements , it s not like intel is manufacturing their own lithography
machinery , they have the same supplier , why would it cost them
less if ever they went a triple patterning route.?.
Do you think that ASML will give them a supplement for free.?.

 

[Abwx]

lol this is too much. Abwx tells us more tales. Maybe the tale where amd magically usurps ipc leadership or where tsm actually gives us a real shrink btwn 2014-2017. Tell us about how magical fdsoi is and how the common platform is the best

Since you have no technical point to add to the discussion
you have to resort to ad hominem attacks , generaly that s
the usual weapon when clulessness is faced with the slightest
tech argument....

 

[liahos1]

Since you have no technical point to add to the discussion
you have to resort to ad hominem attacks , generaly that s
the usual weapon for clulessness....

And what may I ask have you added of any value? Other than your random baseless opinions disguised thinly as "technical"

How about a link or something. Or calculation.

 

[Arachnotronic]

You re looking too much at the PR side of thing and i m not interested
by marketing related material , actualy you re understimating thousands
of scientists and engineers that have full understanding of processes
and subsequent devices caracteristics as well as manufacturing tools
requirements , it s not like intel is manufacturing their own lithography
machinery , they have the same supplier , why would it cost them
less if ever they went a triple patterning route.?.
Do you think that ASML will give them a supplement for free.?.

Intel owns 15% of ASML and is bankrolling a good amount of its R&D for EUV. Pretty sure Intel's getting some preferential treatment.

 

[Abwx]

And what may I ask have you added of any value? Other than your random baseless opinions disguised thinly as "technical"

I keep the discussion at a level wich can be understood by anybody.

Since we re talking Finfets just imagine that thoses transistors
were first used in discrete form in the early 70s as multi gates
mosfets in the high frequency part (800-900MhZ) of TV sets.

Sure that getting the same devices at nm size is quite
a challenge but still their dynamic caracteristics as well
as drawbacks and advantages are documented for a long
time and it happens that i have some knowledge of the
technology in the area that interest us.

 

[Abwx]

Intel owns 15% of ASML and is bankrolling a good amount of its R&D for EUV. Pretty sure Intel's getting some preferential treatment.

They will get 15% of ASML net earnings...and for ASML
to make a profit it s necessary to have enough margin
left if they want to keep pace on increasingly inflated
costs so dont expect them to fire price their machinery.

Besides using a share holder position to get lower prices
from a corporate is simply illegal , the conflict of interests
is just too obvious.