Intel Cannonlake 10 nm delayed, introducing KabyLake

[Fjodor2001]

The desktop refreshes with few to no LGA for broadwell and cannonlake were planned a while ago I think, or at least there were rumors. Only one I could find after a bit of searching was this:
http://semiaccurate.com/2012/11/26/intel-kills-off-the-desktop-pcs-go-with-it/

Is it known what Cannonlake SKUs there will be, and that it'll be few LGA models? I could not find any info on that in the article you linked to.

Also, I think the main question was whether Intel had planned for Haswell Refresh and Kaby Lake well in advance, or if they were decided on ad-hoc and in panic because 14 and 10 nm got delayed.

 

[IntelUser2000]

We are talking about R7 250 here, not 250x. R7 250 has only 384 shaders vs 320 for R7 240, so it will be barely faster. And according to the review at Toms hardware, HD6200 is almost twice as fast at the 240. So even if you allow for toms trying to show iris in a good light, you are way off base and I stand by what I said.

Why do you believe that Toms bench? It uses super low settings. HL2: LC don't you see a problem there alone? No wonder its fast, even with the Iris Pro its running at CPU-bound settings. On average Iris Pro is ~20% faster than best AMD iGPU, not 60%+ as Toms. Don't you find it funny in one benchmark they are using discrete GPUs, its running 1280x720 at minimum settings?

Plus I said R7 240 DDR3. It's barely faster than that.

Iris Pro 6200 only looks good because the reviewers refuse to include 5200 in the comparison. And people never bothered to look at few versus AMD comparisons to see how well the 5200 did already. Newsflash, 5200 was already beating AMD, don't you think the 6200 will?

Update: Benchmarks on the internet show the difference between R7 250 GDDR5 and R7 240 DDR3 as being 50-70%, with some high resolution tests reaching 80-90% range. You'd need 50% gain with Skylake GT4e to be anywhere near R7 250 GDDR5.

 

[frozentundra123456]

Why do you believe that Toms bench? It uses super low settings. HL2: LC don't you see a problem there alone? No wonder its fast, even with the Iris Pro its running at CPU-bound settings. On average Iris Pro is ~20% faster than best AMD iGPU, not 60%+ as Toms. Don't you find it funny in one benchmark they are using discrete GPUs, its running 1280x720 at minimum settings?

Plus I said R7 240 DDR3. It's barely faster than that.

Iris Pro 6200 only looks good because the reviewers refuse to include 5200 in the comparison. And people never bothered to look at few versus AMD comparisons to see how well the 5200 did already. Newsflash, 5200 was already beating AMD, don't you think the 6200 will?

Update: Benchmarks on the internet show the difference between R7 250 GDDR5 and R7 240 DDR3 as being 50-70%, with some high resolution tests reaching 80-90% range. You'd need 50% gain with Skylake GT4e to be anywhere near R7 250 GDDR5.

Why should I not believe the data of Tom's? Just because you dont like the results?

In any any case, I was referring to the post of Aten Ra using data from R7-250 DDR3. It is very close to Kaveri, and by your own admission 6200 is faster than Kaveri. So by logical deduction 6200 will be faster than R7-250 DDR3. Again by your own data, even if R7-250 is 50% faster than the DDR3 version, which you did not document, simple math wil show that 6200 should be about 30% slower than R7-250 GDDR5, which I already consider "somewhere near". If Skylake gt4e is 50% faster, as predicted, it should beat R7-250 GDDR5.

The problem with iris pro is that it uses a lot of die area, and is basically using the process advantage to throw brute force transistors at the performance. And I never said it was or was not faster that 5200. There is plenty to criticize about iris pro, you dont have to make stuff up or shift the goalposts if you want to criticize it.

Edit: The Tom's test at 1080P low shows iris pro already faster than R7-250 GDDR5. Now you can argue with the settings, but even at higher image quality, I dont accept that gt3e, much less 4e will not be "close" to R7-250.

 

[Fjodor2001]

Assuming Intel's 10 nm is delayed to "sometime in 2017" as mentioned in the OP, will this be the crossover point where Intel loses the process tech crown to the competition?

If TSMC and Samsung deliver according to plan, they should have 10 nm on the market earlier than Intel, right?

 

[soccerballtux]

1500+ balls for a 4.5W chip ???

wow. that takes some balls

 

[Ajay]

Assuming Intel's 10 nm is delayed to "sometime in 2017" as mentioned in the OP, will this be the crossover point where Intel loses the process tech crown to the competition?

If TSMC and Samsung deliver according to plan, they should have 10 nm on the market earlier than Intel, right?

Well, as IDontCare has probably said - not all 10 nm nodes are the same. We'll just have to see how they stack up.

 

[Phynaz]

Assuming Intel's 10 nm is delayed to "sometime in 2017" as mentioned in the OP, will this be the crossover point where Intel loses the process tech crown to the competition?

If TSMC and Samsung deliver according to plan, they should have 10 nm on the market earlier than Intel, right?

No

 

[mrmt]

Assuming Intel's 10 nm is delayed to "sometime in 2017" as mentioned in the OP, will this be the crossover point where Intel loses the process tech crown to the competition?

If TSMC and Samsung deliver according to plan, they should have 10 nm on the market earlier than Intel, right?

Yes, and with Jim Keller ahead of the Zen family, AMD will crush Intel core to the bone. :awe:

 

[DrMrLordX]

wow. that takes some balls

Great minds think alike.

 

[PaulIntellini]

with the removal of FIVR, you need more pins for power delivery.
BDW-Y has 1234 pins.

secondly, the Y models have tighter ball pitch for improved density of signal pins.
but then you need more pins for power delivery.

 

[Fjodor2001]

No

Thanks for your elaborative and insightful response. As always.

 

[Fjodor2001]

Yes, and with Jim Keller ahead of the Zen family, AMD will crush Intel core to the bone. :awe:

It's not AMD we're talking about, but TSMC and Samsung and process tech. I.e. something completely different.

Do you have some evidence contradicting that they will deliver 10nm ahead of Intel if they deliver according to plan? Or were you just thread crapping?

 

[Fjodor2001]

Well, as IDontCare has probably said - not all 10 nm nodes are the same. We'll just have to see how they stack up.

Yep, there are different pros and cons of each process tech, and it differs per metric. Summing it all up I get the impression that Intel's 10 nm in general has an advantage, but it's debatable by how much. It doesn't look like there is any consensus yet on that issue on all aspects, and not all technical details are known either.

But anyway, from what is known at the moment, will TSCM and Samsung deliver their version of 10 nm before Intel delivers their 10 version of nm?

 

[ShintaiDK]

Yep, there are different pros and cons of each process tech, and it differs per metric. Summing it all up I get the impression that Intel's 10 nm in general has an advantage, but it's debatable by how much. It doesn't look like there is any consensus yet on that issue on all aspects, and not all technical details are known either.

But anyway, from what is known at the moment, will TSCM and Samsung deliver their version of 10 nm before Intel delivers their 10 version of nm?

Do I need to remind you?

TSMC did nothing but rename a product with added FF for example. TSMC and Samsungs 10nm may be what Intels 14nm is.

Not to mention we all know how and when TSMC tends to deliver. And Samsung is so eager to show off they forget to be able to deliver volume. Hence Apple is forced to source from 3 foundries for A9/A9X. That basicly means all capacity from 3 foundries more or less goes to Apple in the start.

 

[Nothingness]

Hence Apple is forced to source from 3 foundries for A9/A9X.

Is there any official source for that (of course that wouldn't be from Apple :biggrin🙂? The development costs (implementation takes a *lot* of work) and production costs (masks) will explode if they have to use 3 foundries.

 

[Fjodor2001]

Do I need to remind you?

TSMC did nothing but rename a product with added FF for example. TSMC and Samsungs 10nm may be what Intels 14nm is.

Not to mention we all know how and when TSMC tends to deliver. And Samsung is so eager to show off they forget to be able to deliver volume. Hence Apple is forced to source from 3 foundries for A9/A9X. That basicly means all capacity from 3 foundries more or less goes to Apple in the start.

We're talking about 10 nm, not 14 nm. Got any data on that? And what about all other metrics - power consumption, actual transistor density, etc?

Also, what does it matter who buys the TSMC/Samsung 10 nm wafers initially. They are still on the market, right?

 

[ShintaiDK]

Is there any official source for that (of course that wouldn't be from Apple :biggrin🙂? The development costs (implementation takes a *lot* of work) and production costs (masks) will explode if they have to use 3 foundries.

http://appleinsider.com/articles/15...tsmc-for-30-of-a9-chip-orders-for-next-iphone

A9 is to be made at Samsung, GloFo and TSMC.
A9X to be made at TSMC.

GloFos 30% yield with a 100mm2 chip cant surprise anyone 😛

 

[Nothingness]

http://appleinsider.com/articles/15...tsmc-for-30-of-a9-chip-orders-for-next-iphone

A9 is to be made at Samsung, GloFo and TSMC.
A9X to be made at TSMC.

GloFos 30% yield with a 100mm2 chip cant surprise anyone 😛

Yeah I know that source, but it certainly isn't official 😉 I guess we'll have to wait for the real thing to see.

 

[ShintaiDK]

Yeah I know that source, but it certainly isn't official 😉 I guess we'll have to wait for the real thing to see.

I think its quite (un)official that TSMC makes A9 now.

http://www.phonearena.com/news/Report-TSMC-to-start-Apple-A9-mass-production-this-month_id70312

 

[mrmt]

It's not AMD we're talking about, but TSMC and Samsung and process tech. I.e. something completely different.

Do you have some evidence contradicting that they will deliver 10nm ahead of Intel if they deliver according to plan? Or were you just thread crapping?

I'm just going a bit ahead of you and reaching the right conclusions you are likely to put in other threads.

As for the question regarding TSMC/Samsung 10nm, it doesn't really matter if they are going the first to reach 10nm on the market, what they are going to deliver calling 10nm is what matter and that's what makes your question largely irrelevant. I can't even think why TSMC and Samsung don't rename their 10nm process a 7nm process and call victory over Intel. Why only rename to reach parity like they did?

 

[carop]

The development costs (implementation takes a *lot* of work) and production costs (masks) will explode if they have to use 3 foundries.

GlobalFoundries and Samsung are GDSII compatible. So, It is possible to use a single PDK to carry out a single design, and send a single GDSII file to either company.

They are not, however, mask compatible:

However, customers will have to re-do lithography masks if they want to move manufacturing from one company to the other, in part because of issues with shipping masks.

http://semimd.com/blog/2014/04/17/globalfoundries-and-samsung-join-forces-on-14nm-finfets/

 

[Fjodor2001]

I'm just going a bit ahead of you and reaching the right conclusions you are likely to put in other threads.

As for the question regarding TSMC/Samsung 10nm, it doesn't really matter if they are going the first to reach 10nm on the market, what they are going to deliver calling 10nm is what matter and that's what makes your question largely irrelevant. I can't even think why TSMC and Samsung don't rename their 10nm process a 7nm process and call victory over Intel. Why only rename to reach parity like they did?

That has already been discussed, and there is no consensus on all aspects of it. Two different process techs will never be identical on all metrics. And their certainly is no consensus that TSMC/Samsung 10 nm should be same or worse than Intel 14 nm like you claim. How close TSMC/Samsung 10 nm is to Intel 10 nm can be discussed though, and we won't know for certain for some time. In the end it will differ per metric compared.

Anyway, since you refused to answer the actual question, I assume you agree that TSMC/Samsung will be releasing their version of 10 nm before Intel releases their version of 10 nm. Intel's 10 nm may be a bit better, but if it also is released later we might have to call it a tie in the end. In that case it's a huge loss in process tech lead for Intel compared to how things were a couple of years ago, no matter how you try to spin it. Or would you not agree with that?

 

[Phynaz]

Thanks for your elaborative and insightful response. As always.

You asked a yes or no question. Perhaps the issue is with the question and not the answer.

We already know you are an Intel hater and you're trying to get somebody to give you joy over this rumor of a delay.

Here's the fact: TSMC will be on 28nm for over 5 years before they have another high power node ready. Intel's manufacturing lead isn't shrinking, it's growing.

 

[NTMBK]

Here's the fact: TSMC will be on 28nm for over 5 years before they have another high power node ready. Intel's manufacturing lead isn't shrinking, it's growing.

The fact is that the big bucks aren't in high performance, they're in mobile. Hence 20nm being mobile only, and 14nm coming to mobile first.

 

[mrmt]

That has already been discussed, and there is no consensus on all aspects of it. Two different process techs will never be identical on all metrics. And their certainly is no consensus that TSMC/Samsung 10 nm should be same or worse than Intel 14 nm like you claim. How close TSMC/Samsung 10 nm is to Intel 10 nm can be discussed though, and we won't know for certain for some time. In the end it will differ per metric compared.

First and foremost consensus isn't needed to achieve what you want, plus if you are admiting by yourself that the technologies under the same marketing name aren't the same, what's the point in comparing them? Better to ask "when will manufacturer XYZ release a node with the following parameters" than using a meaningless marketing moniker.