Intel Skylake / Kaby Lake

[ThatBuzzkiller]

Those cost/transistor charts are normalized for yield. If your yields aren't like-for-like, then the cost/transistor comparisons are worthless.

I don't believe yields will be a differentiator ...

People argued FD-SOI in favour of FinFETs because of higher yields but that turned out to be a myth ...

TSMC has near ideal yields for 20nm node yet their not touting cost reduction benefits like Intel is and neither is Samsung with their 14nm node but they all have fairly state of the art photolithography technology ...

Yields appear to be very easy to ramp up in comparison to lowering masking costs and it's not like the chip designers couldn't account for more redundancy to increase the yields at a small price ...

How exactly can Intel claim that they are getting a lower cost per transistor than they were able to before when their wafer costs are spiraling out of control much like the other IDMs ?

It's small things like these that make me skeptical of Intel's claim on cost reduction benefits when they've likely yet to solve the masking costs ...

Professionals like Cliff Maier and Bob Colwell don't seem to believe in Intel either and their both microprocessor designers BTW ...

 

[ThatBuzzkiller]

Because they can afford the design for it (Cost(due to volume) and time). Just read the article I linked. You will find answer to some of your questions there.

"Affording" it does not mean they've solved the specifics like the masking costs ...

 

[ShintaiDK]

You're off by a factor of 4, Intel are "only" 15% of the semi industry revenue.

http://www.gartner.com/newsroom/id/3023317

Sorry i was thinking MPU revenue. Since that is where you may see leading edge nodes.

 

[ShintaiDK]

"Affording" it does not mean they've solved the specifics like the masking costs ...

Did you read the article? IBS isnt some random nobody in this field.

You would also notice that GloFo/TSMC got very low yield compared to Intel. Even 100mm2 chips today on 20nm got what, 50% yield?

 

[Nothingness]

Sorry i was thinking MPU revenue. Since that is where you may see leading edge nodes.

It makes sense

 

[t0mt0m]

If they're launching at Gamescon, and then giving more at IDF 18th August, would we expect to have seen more leaks/information from Intel, going on past launches of Broadwell, Haswell etc? Does the expected time between announcement and release seems normal?

 

[ThatBuzzkiller]

Did you read the article? IBS isnt some random nobody in this field.

You would also notice that GloFo/TSMC got very low yield compared to Intel. Even 100mm2 chips today on 20nm got what, 50% yield?

Yes I did and no specifics were told about how Intel bucked the trend in terms of wafer costs ...

FYI, that chart is over a year old. By now GF/TSMC should have a yield of over 80% on a 100 mm^2 die. Intel maybe the best but that doesn't mean their on a whole other level compared to other top of the line semiconductor foundries ...

https://www.semiwiki.com/forum/content/3292-handel-jones-fd-soi-vs-finfet.html

As far as IBS been some random nobody, I'm very curious about the CEO's educational background and previous work experience since the industry didn't seem to agree with his assessment judging by the link above ...

Handel Jones strikes me more as businessman rather than an engineer ...

 

[ShintaiDK]

I think you forget gate utilization that depends on design. An example is 14/16nm with almost half the gates wasted. gate utilization dropped significantly with multipatterning.

This is about business.

Else you could get 7nm chips today if it was all about engineering.

Semiwiki isnt the place for neutral information. Its pretty much a TSMC PR site.

 

[Anth Seebel]

doh

 

[ThatBuzzkiller]

I think you forget gate utilization that depends on design. An example is 14/16nm with almost half the gates wasted. gate utilization dropped significantly with multipatterning.

This is about business.

Else you could get 7nm chips today if it was all about engineering.

Semiwiki isnt the place for neutral information. Its pretty much a TSMC PR site.

Gate utilization is mostly dependent on how good the design tools are and I don't see TSMC or Samsung faltering in this area seeing as how every chip designer is investing more on it too ...

You are right that it's about business but in this case it is ultimately dependent on the engineering aspect ...

I'm not arguing whether or not Semiwiki is a neutral place for information but rather I'm showing that Handel Jones doesn't always have the best insights when it comes to extremely important problems chip manufacturing ...

 

[ShintaiDK]

Tell that to the foundry customers. Everyone expect exactly what IBS projects.

 

[Arachnotronic]

Tell that to the foundry customers. Everyone expect exactly what IBS projects.

Yes, IBS' views have been echoed by many top semiconductor companies.

 

[ThatBuzzkiller]

Tell that to the foundry customers. Everyone expect exactly what IBS projects.

Everyone ?

They projected that FD-SOI would be cheaper than FinFets yet it's strange how no IDMs are taking the initiative to capitalize on it ...

They are also projecting 10nm to be cheaper than 20nm or 14nm per transistor but we'll see soon enough how many chip designers bite on it despite the fact that neither TSMC or Samsung are touting cost advantages at that point ...

I'm not interested in whatever IBS says, I want to know how Intel can keep claiming cost reduction advantage and badly ...

Notice how their only real big customer is Altera and how they sell their FPGAs at very high margins ...

Design cost are only a small part of the equation in comparison to the skyrocketing wafer costs so they should more than cancel out but don't you think it's strange how no chip designers are desperate enough to look at Intel ...

Even though some of them are competitors to some of Intel's products they could still be making big bucks from the likes Apple or Nvidia if they were manufacturing chips for them plus Intel would have a monopoly on chip manufacturing which effectively means locking out other IDMs from doing high end investments to compete with Intel ...

 

[Kallogan]

skylake will be the first time i will consider buyin a laptop without dgpu

iris pro 6200 is already quite beasty

not only the perfs are great but the power comsumption is insanely low :

http://www.tomshardware.fr/articles/intel-broadwell-core-i7-5775c,2-2384-3.html

48 EUs, 72 will rock it

 

[mikk]

Compute Performance of Intel® Core™ i5-6200U CPU with HD Graphics 520
https://compubench.com/device.jsp?b...R) Core(TM) i5-6200U CPU with HD Graphics 520


Although we don't know system details like memory, here is a Broadwell ULT i5.


Compute Performance of Intel® Core™ i5-5300U CPU with HD Graphics 5500
https://compubench.com/device.jsp?b...) Core(TM) i5-5300U CPU with HD Graphics 5500

 

[Fjodor2001]

Seems like Iris Pro is a very expensive band aid to solve Intel's GPU disadvantage compared to the competition.

 

[LTC8K6]

Seems like Iris Pro is a very expensive band aid to solve Intel's GPU disadvantage compared to the competition.

Doesn't look any more expensive than before, and it's a lot faster.

Hopefully the full GT4e with Skylake will be impressive for an iGPU.

It's certainly going to clean HD4600's clock. :biggrin:

 

[Sweepr]

Seems like Iris Pro is a very expensive band aid to solve Intel's GPU disadvantage compared to the competition.

If they wanted to match AMD a smaller 2C/4T Broadwell GT3e would be more than enough. Quad-core Broadwell-K is simply a superior chip from a CPU+iGPU perspective. Skylake GT4e and Kabylake GT4e 2x 128MB gaming performance will only widen the gap next year.
Despite limited availability, the point is Intel has the performance lead on iGPUs and an agressive 2015/2016 roadmap ahead.

 

[Fjodor2001]

Certainly not when perf/price is taken into account. AMD has the lead there.

 

[Phynaz]

Seems like Iris Pro is a very expensive band aid to solve Intel's GPU disadvantage compared to the competition.

Bandaid? Is that what it is when they have the performance lead?

Will you be calling Zen a bandaid?

 

[Phynaz]

Certainly not when perf/price is taken into account. AMD has the lead there.

Not really. It all depends on how you measure it. Does price mean TCO? Is power consumption a performance parameter? Heat output? How about CPU performance?

 

[PPB]

Bandaid? Is that what it is when they have the performance lead?

Will you be calling Zen a bandaid?

This is a non-sequitur. Zen is a product aimed to be sported in almost every SKU AMD plans to sell from the end of 2016 onwards, Desktop Broadwell is just a tiny niche product made probably to save face considering the general 14nm delays, the focus of Broadwell into the mobile space and the lack of a iris-pro driven Broadwell SKU to supercede the Haswell based one. While Zen is spammed in almost the AMD's entire roadmap, Iris pro desktop Broadwell just takes a tiny spot in Intel's. Not even mentioning the difference in value between CPU and iGP in a x86 APU. Yours is just another unfounded fallacy.

 

[frozentundra123456]

It certainly is a non-sequitor. Broadwell actually exists. Zen is just a powerpoint slide for at least another year.

 

[Fjodor2001]

Bandaid? Is that what it is when they have the performance lead?

Will you be calling Zen a bandaid?

Iris Pro is a bandaid solution because Intel has a worse GPU arch than AMD, so they have to compensate by using more GPU die area plus expensive eDRAM.

 

[CHADBOGA]

This is a non-sequitur. Zen is a product aimed to be sported in almost every SKU AMD plans to sell from the end of 2016 onwards, Desktop Broadwell is just a tiny niche product made probably to save face considering the general 14nm delays, the focus of Broadwell into the mobile space and the lack of a iris-pro driven Broadwell SKU to supercede the Haswell based one. While Zen is spammed in almost the AMD's entire roadmap, Iris pro desktop Broadwell just takes a tiny spot in Intel's. Not even mentioning the difference in value between CPU and iGP in a x86 APU. Yours is just another unfounded fallacy.

You should be comparing like with like, and compare Zen to Skylake & future Intel products it will be up against.