Speculation: Intel will become fabless

[LTC8K6]

Z370 is 22nm according to Intel.

 

[JDG1980]

But Intel 10nm and everyone elses's 7nm are roughly the same. So saying "already have 7nm out" is misleading. It makes it seem like Intel is a gen behind, when they are really just late on the same gen.
These nanometer terms are largely marketing now.

I hear this stuff all the time, that Intel's processes are better than TSMC/Samsung/GloFo processes with the same "nanometer" rating. But I'm not at all convinced that it is true. If it was, then you'd expect Intel's 14nm CPUs to be denser than 14nm Ryzen CPUs, and to have noticeably better performance per watt. But neither of these is true. Intel no longer publishes die sizes and transistor counts, which is a warning sign in and of itself (they used to readily do so). This comparison indicates that AMD has the edge in terms of cores by die area, at least on HEDT. Cherry-picked numbers published by Intel regarding SRAM density have little relevance if they don't translate into a real-world advantage in CPU density. As for performance per watt, Ryzen stacks up quite well to Coffee Lake, especially in multi-threaded benchmarks.

One area where Intel does seem to have an advantage over AMD is in terms of maximum clock speed. It's not clear to what extent that is due to process, and to what extent it is due to architecture. Ryzen is currently being fabbed on a GloFo 14LPP process, which is optimized for mobile applications; TSMC has high-performance processes available, which will probably provide better clocks when AMD switches to that process for 7nm. If Pinnacle Ridge were currently being fabbed on TSMC 16FF+ instead of GloFo, it's entirely possible it could clock nearly as high as Coffee Lake, so it's not clear that foundry processes are inferior to Intel processes with the same "nanometer" rating in any way.

 

[LTC8K6]

Ryzen 2700/2600 series are on 12nm.

Intel is still on 14nm and doing unbelievably well with it.

 

[PingSpike]

"H310 and several other 300 series desktop processors"... I assume they mean chipsets?

So they are only outsourcing the chipsets then?

I didn't even think they made the chipsets.

Of course they make the chipsets. That's why they change sockets so often, so they can sell a new chipset when some one wants a cpu upgrade!

 

[Kenmitch]

Maybe a start?

https://www.techpowerup.com/247488/tsmc-to-build-intel-14nm-processors-and-chipsets

 

[LTC8K6]

Maybe a start?

https://www.techpowerup.com/247488/tsmc-to-build-intel-14nm-processors-and-chipsets

Post #74 article seems to say it's only chipsets.

https://www.digitimes.com/news/a20180910PD210.html

 

[LTC8K6]

Intel denies TSMC CPU outsourcing, relying on its own investment to cope with unexpected 14nm demand

https://www.pcgamesn.com/intel-denies-tsmc-outsourcing

But we have just received an official statement from Intel itself, which says: “In response to the stronger than expected demand environment, we are continuing to invest in Intel’s 14nm manufacturing capacity.”

That seems to make it pretty clear that it’s not going to be looking for help outside of the company and will be working its own facilities hard to make up the shortfall in its silicon stockpile.

 

[NTMBK]

Intel denies TSMC CPU outsourcing, relying on its own investment to cope with unexpected 14nm demand

https://www.pcgamesn.com/intel-denies-tsmc-outsourcing

They seem to be reading an awful lot into that one line. They're investing in Intel's 14nm manufacturing, sure... but that doesn't mean that they can't also be fabbing some chips at TSMC in the short term.

 

[iBoMbY]

This is not a denial for outsourcing to TSMC, just a deflection:

In response to the stronger than expected demand environment, we are continuing to invest in Intel’s 14 nm manufacturing capacity.

 

[NostaSeronx]

Intel is most likely not going to port CPU chips to TSMC.

Intel's 12nm is soon. 10nm FinFET on the 14nm/14nm+ design rules. Implemented as "Beyond 14nm" node.
The Arizona Fab is starting 7nm with/on Nikon NSR-S635E in 2H2019.

14nm++/14nm+++ CPP=84nm -> 12nm/12nm+ CPP=70nm(64nm in HP)
7nm is a shrink of 10nm and is expected to enter much faster and hit HVM yields sooner than 10nm+.

 

[LTC8K6]

This is not a denial for outsourcing to TSMC, just a deflection:

It was never CPU outsourcing though, just chipsets. And Z370 is 22nm anyway, so the story sounds odd.

 

[jpiniero]

It was never CPU outsourcing though, just chipsets. And Z370 is 22nm anyway, so the story sounds odd.

H310 is way more volume than z series.

 

[tynopik]

it seems like each die shrink is gaining less and less

does being a generation behind in process tech really matter in terms of fielding a competitive cpu/gpu?

 

[LTC8K6]

H310 is way more volume than z series.

Digitimes earlier said H310 production was stopped, purportedly to change it to 22nm.

https://www.digitimes.com/news/a20180508PD211.html

 

[moinmoin]

does being a generation behind in process tech really matter in terms of fielding a competitive cpu/gpu?

For seemingly ages it was Intel ahead of the rest of the industry.

Currently it's starting to change to everybody using TSMC ahead of Intel. We will see whether or not that matters to Intel and its competitiveness.

 

[jpiniero]

Digitimes earlier said H310 production was stopped, purportedly to change it to 22nm.

https://www.digitimes.com/news/a20180508PD211.html

Could be the kind of thing that they realized they needed a fab to take over H310 production completely and not just temporarily and don't have the 22 nm capacity anymore since it was reduced to make room for more 14.

 

[IntelUser2000]

Intel is reportedly the only vendor for Apple's Gigabit LTE modems on the new iPhones.

iPhones sell 200 million a year! XMM 7560 is the Intel modem that'll be going in this generation. XMM 7560 is on 14nm.

Apple has so much influence on its suppliers because of the sheer amount of phones it ships.

 

[DrMrLordX]

does being a generation behind in process tech really matter in terms of fielding a competitive cpu/gpu?

Yes. It's going to hurt Intel in the server room the most, though it may also hurt them in the mobile PC market.

There's only so much they can do without node shrinks. They were supposed to crap out somewhere in the 3-5nm range, and instead Intel seems to have run aground at 14nm.

 

[scannall]

Yes. It's going to hurt Intel in the server room the most, though it may also hurt them in the mobile PC market.

There's only so much they can do without node shrinks. They were supposed to crap out somewhere in the 3-5nm range, and instead Intel seems to have run aground at 14nm.

It won't hurt them much in mobile, at least not right away. They have solid brand recognition and will sell a ton of 'Mom Boxes" just on name. In servers? That may hurt them over time. Server space moves slowly, and doesn't change direction on a dime. But once it moves away from you, it takes a long time to head back your way. Meltdown will hasten the turn.

 

[ITSTours]

So this thread is more than a year old... let's revisit this again.

Intel still doesn't have a 10nm desktop chip. There are a few mobile chips only. The base clock is kinda abysmal.
There's a rumor that Intel will completely skip 10nm and go onto 7nm for desktop chips. (Intel has denied.)

Now what do you think, guys? Does Intel need to go fabless?
Or will it magically return with an awsum 7nm node that actually can compete with TSMC/Samsung 5nm?

 

[jpiniero]

Now what do you think, guys? Does Intel need to go fabless?
Or will it magically return with an awsum 7nm node that actually can compete with TSMC/Samsung 5nm?

If 7nm is a bust like 10 nm, Intel will have no choice, just simply because they won't be able to afford any further node.

 

[scannall]

They may have to come up with a real foundry business model, instead of fiddling around the edges. Spread some of the R&D cost around, so they don't have to eat it all.

 

[insertcarehere]

If 7nm is a bust like 10 nm, Intel will have no choice, just simply because they won't be able to afford any further node.

If Intel actually throws in the towel and goes fabless, TSMC and Samsung will have even more mind-boggling amounts of power to decide the fates of many of the world's largest tech companies, pretty scary to think about, really.

 

[thesmokingman]

If they went fabless, the world would slow to a crawl. That's how much volume they handle. The top 5 fabs would be so overloaded no one would get their phones or chips.

 

[jpiniero]

Hmm.. Swan said that they expect the 7 nm GPGPU in Q4 2021. If they are only expecting to be what sounds like the equivalent of the foundries' risk production ability/levels, that makes it tough to think they will be able to move that quickly to bring up other products other than to satisfy Aurora's needs. Kind of rough when you consider that TSMC is saying they expect to go volume production with 5 nm in a couple months.