Intel Rumored to be Courting GlobalFoundries for Some CPU Manufacturing

[btarunr]

With its own silicon fabrication facilities pushed to their capacity limits, Intel is looking for third-party semiconductor foundries to share some of its supply load, and according to a WCCFTech report, its latest partner could be GlobalFoundries, which has a 14 nm-class fab in Upstate New York. If it goes through, the possible Intel-GloFo deal could see contract manufacturing commence within 2020.

GloFo's fab offers 14 nm FinFET and 12LPP, a refinement that's marketed as 12 nm. According to the report, Intel could use GloFo for manufacturing CPU dies, specifically its entry-level chips such as Core i3, Pentium, and Celeron. Intel is also known to shed its own manufacturing workload by contracting foundries for 14 nm core-logic (chipsets). In a bid to maximize 14 nm fab allocation for its CPUs, Intel also started making some of its 300-series chipsets on the older 22 nm process, which goes to show the company's appetite for 14 nm.



View at TechPowerUp Main Site

 

[fancucker]

Backported Willow Cove will still be very competitive against Zen 3. In typical fashion the red team can only equal or beat with a massive node advantage.

EMIB based multi-chiplet 14nm server dies will still beat them, and AMD can't touch them in AVX512 and its subset instructions

 

[btarunr]

Backported Willow Cove will still be very competitive against Zen 3. In typical fashion the red team can only equal or beat with a massive node advantage.

EMIB based multi-chiplet 14nm server dies will still beat them, and AMD can't touch them in AVX512 and its subset instructions

Except that Rocket Lake caps out at 8 cores per socket. At best these will be good gaming CPUs, given that 8-core/16-thread will be plenty for gaming even in 2021.

 

[ZoneDymo]

Backported Willow Cove will still be very competitive against Zen 3. In typical fashion the red team can only equal or beat with a massive node advantage.

EMIB based multi-chiplet 14nm server dies will still beat them, and AMD can't touch them in AVX512 and its subset instructions

imagine still being positive about Intel in 2020

 

[Crackong]

Intel (AMD Inside) -Inside ?

 

[R-T-B]

Ah. East fishkill. I remember that plant from Ryzen 1, and when it made PPC970 chips for Apple.

Honestly, I have nostalgia for the place for some weird reason, but I doubt this'll really happen.

 

[oxidized]

imagine still being positive about Intel in 2020

Imagine being a fanboy of manufacturers of PC components in 2020.

 

[windwhirl]

Imagine being a fanboy of manufacturers of PC components in 2020.

Or at any time.

 

[Valantar]

I doubt this is true, but there would certainly be a particularly odd sort of irony if true. Good on GloFo, I suppose Also about time Intel takes some drastic measures to alleviate their supply shortages.

Backported Willow Cove will still be very competitive against Zen 3. In typical fashion the red team can only equal or beat with a massive node advantage.

EMIB based multi-chiplet 14nm server dies will still beat them, and AMD can't touch them in AVX512 and its subset instructions

Very competitive? Probably in performance, not in power efficiency. GloFo 14nm is less efficient than Intel 14nm, and Zen2 is matching or beating Sunny Cove on 10nm (final verdict waiting on Renoir reviews, but 8 7nm Zen2 cores at 1.8GHz base at 15W vs 4 10nm SC cores at 1.3GHz base at 15W does say something about relative efficiency nonetheless. Even if the Zen part consumes 2x the power when running at base clocks fully loaded it's still more efficient, as SC's IPC advantage over Zen2 is only about 10% (Zen2 is about 6.5% ahead of CFL IPC according to Anandtech's testing in SPEC2017)). Any backported Willow Cove will struggle to compete.

As for AVX512, the use cases for normal people are still far too limited for this to actually matter. For now. Might change in five years, but by then AMD will have their own version.

 

[ratirt]

AMD ditches GloFo and goes with TSMC and now Intel is going GloFo. What an interesting change.

 

[I No]

WCCFTech .... a truckload of salt .... the only thing I'd trust out of that place is an article that they copied off a decent website.

 

[ZoneDymo]

Imagine being a fanboy of manufacturers of PC components in 2020.

or a fanboy of anything lets be honest, makes no sense.

 

[95Viper]

Stay on topic and stop the fanboy discussions/remarks.
No insulting others.

Thank You and Have a Good Morning.

 

[notb]

Totally sensible and IMO probable.
Intel will look for more and more possibilities. GF is cheap - perfect for the entry level models.

AMD ditches GloFo and goes with TSMC and now Intel is going GloFo. What an interesting change.

The whole Zen idea is built around being on the best node available to stay competitive.
People keep saying that the common die means big savings during binning - virtually any Zen die can be used somehow if it has 2+ working cores.
What they forget (or ignore) is that AMD has to make all dies using a very expensive process.

Intel makes many different designs on different nodes, but it lets them make the entry-level stuff with minimal cost in the first place.

The bottom line is that - at least for now - Intel's gross margin remains much higher.

 

[Valantar]

Totally sensible and IMO probable.
Intel will look for more and more possibilities. GF is cheap - perfect for the entry level models.


The whole Zen idea is built around being on the best node available to stay competitive.
People keep saying that the common die means big savings during binning - virtually any Zen die can be used somehow if it has 2+ working cores.
What they forget (or ignore) is that AMD has to make all dies using a very expensive process.

Intel makes many different designs on different nodes, but it lets them make the entry-level stuff with minimal cost in the first place.

The bottom line is that - at least for now - Intel's gross margin remains much higher.

Intel's gross margin can largely (though not entirely) be attributed to their massive market share in the enormously profitable enterprise/server/datacenter/HPC space. Even with costly service contracts and massive rebates off "list prices" for anyone buying at scale, it's still the same silicon that they sell to consumers (either as regular desktop chips or HEDT ones) at massive markups. Intel does not have much of a history of producing low end chips (outside of Atom) on cheaper nodes than the rest. And consumer parts always have low margins, which is why they are focusing on maintaining supply only for the more profitable higher priced ones.

I still agree that this sounds plausible and smart, though. On the other hand one has to wonder about the performance and power efficiency of these parts. It would also be incredibly fascinating to compare such a part to an AMD part made on the same node. That's as like-for-like a comparison of architectural performance and efficiency as you can get. Has this ever been possible before?