GLOBALFOUNDRIES Introduces New 12nm FinFET Technology for High-Performance Applications

[Ajay]

I still don't see how it adds up. Let's say GloFo makes it to risk production in January. AMD gets first silicon back on January 2nd - the very first wafer off the line. How do you qualify that silicon, place a wafer order, get the wafers manufactured, cut, tested, packaged and shipped in 30 - 60 days.

They don't. Either AMD is really going to be producing in 4Q17 or they aren't going to be shipping anything in February 2018. This upcoming Digitimes article ought to be interesting - unless they are starting to try click-bait articles.

 

[CatMerc]

They don't. Either AMD is really going to be producing in 4Q17 or they aren't going to be shipping anything in February 2018. This upcoming Digitimes article ought to be interesting - unless they are starting to try click-bait articles.

Yeah, agreed.

Possibility: They're launching 12nm chips NVIDIA Pascal style: A month before actual product is available.

 

[Abwx]

https://www.globalfoundries.com/new...-technology-for-high-performance-applications

Perfs improvements are relative to competing solutions not to GF s current 14nm :

The new 12LP technology provides as much as a 15 percent improvement in circuit density and more than a 10 percent improvement in performance over 16/14nm FinFET solutions on the market today.

Also there s two 12nm processes, the wording suggest that their FD-SOI offering, 12FDXTM, use planar transistors :

GF's new 12nm FinFET technology complements its existing 12nm FD-SOI offering, 12FDXTM. While some applications require the unsurpassed performance of FinFET transistors, many connected devices need high levels of integration and more flexibility for performance and power consumption, at costs FinFET cannot achieve. 12FDX provides an alternative path for the next generation of connected intelligent systems, enabling the performance of 10nm FinFET with better power consumption, lower cost, and better RF integration than current-generation foundry FinFET offerings.

 

[coercitiv]

From DigiTimes as Arachnotronic already said it was coming:

AMD has informed its partners that it plans to launch in February 2018 an upgrade version of its Ryzen series processors built using a 12nm low-power (12LP) process at Globalfoundries, according to sources at motherboard makers.

The company will initially release the CPUs codenamed Pinnacle 7, followed by mid-range Pinnacle 5 and entry-level Pinnacle 3 processors in March 2018, the sources disclosed. AMD is also expected to see its share of the desktop CPU market return to 30% in the first half of 2018.

Their corresponding chipsets, the 400 series, will also become available in March 2018 with X470- or B450-based motherboards to be the first to hit the store shelves.

 

[Arachnotronic]

From DigiTimes as Arachnotronic already said it was coming:

That's really good execution from AMD if they can hit this schedule.

 

[R0H1T]

Why do they need a new set of chipsets?

 

[ajc9988]

So here is a leaked roadmap:
https://hothardware.com/news/leaked-amd-roadmap-zen-2-matisse-12nm-finfet-cpu-2019

 

[CatMerc]

That's really good execution from AMD if they can hit this schedule.

I expected AMD to be aggressive, but that's a level I didn't expect from them. If it actually works out, that's just another example of how different a company AMD has become in recent years.

 

[ajc9988]

Why do they need a new set of chipsets?

To get ready for DDR5 next summer or introduce the pcie 4.0 standard early.

 

[R0H1T]

To get ready for DDR5 next summer or introduce the pcie 4.0 standard early.

DDR5 is still at least a year or two away, unless Intel is also looking to aggressively upgrade their 10nm products with it.
As for pcie 4.0 lanes that's from the CPU anyway, so does that mean PR already has it?

 

[ajc9988]

DDR5 is still at least a year or two away, unless Intel is also looking to aggressively upgrade their 10nm products with it.
As for pcie 4.0 lanes that's from the CPU anyway, so dops that mean PR already has it?

Actually, the DDR5 standard is published next year, summer of 2018. GF 7LP incorporates support for DDR5. Because of this, it could be preparing for the future.

Another alternative is they are trying to help board partners sell more boards, as Intel does, but without cutting off generational compatibility, in an attempt to curry more favour and support from board manufacturers.

Another alternative is changes for Znand or licensing Optane.

As to it being from the cpu, it is easy to make it clear what minimum specs a board supports by creating a chipset revision. This can become a shorthand for features, such as pcie 4.0 (or 5.0, published Q1 2019) or DDR5, thereby aiding consumers.

Now, regardless of consumer availability, do you remember when Intel, years back, added dual compatibility to its HEDT line. Something similar could happen here.

In other words, many possibilities, no certainty.

 

[maddie]

I expected AMD to be aggressive, but that's a level I didn't expect from them. If it actually works out, that's just another example of how different a company AMD has become in recent years.

Didn't Papermaster state this very clearly. An annual refresh/upgrade for at least 4 years. This was put in place yrs ago and seems to be following schedule. The 7LP step is the one that can introduce delays IMO.

 

[formulav8]

Why do they need a new set of chipsets?

There are reasons to bring out a new chipset. Bug fixes, Thunderbolt becomes royalty free in a few months IIRC, so they could add that feature. Maybe their own audio solution like I think Intel did with z370 and maybe possible to add more pcie lanes to add other options/features. Of course money.

PRidge will stay backwards compatible with current chipsets as well.

 

[raghu78]

That's really good execution from AMD if they can hit this schedule.

yeah. The question remains as to how much of a clock improvement can AMD get with Pinnacle Ridge. If AMD can hit 4 Ghz base and 4.5 Ghz max turbo with avg max OC of 4.6-4.7 Ghz they will have a very competitive product. So it all boils down to how good the 12LP process is and has AMD made improvements to the Zen physical design to extract any more clock speed headroom.

 

[CatMerc]

Didn't Papermaster state this very clearly. An annual refresh/upgrade for at least 4 years. This was put in place yrs ago and seems to be following schedule. The 7LP step is the one that can introduce delays IMO.

I expected April-June refresh.

 

[rainy]

I think, that 4 GHz base clock should be achievable with that improved 14nm (aka 12nm), however I'm expecting some IPC gains too:

When asked how long Zen would last, compared to Intel’s two-year tick-tock cadence, Papermaster confirmed the four-year lifespan and tapped the table in front of him: “We’re not going tick-tock,” he said. “Zen is going to be tock, tock, tock.”

https://www.pcworld.com/article/315...hitecture-is-expected-to-last-four-years.html

I expected April-June refresh.

AMD must be aggressive, therefore I'm expecting one year span (March-April).

 

[Ajay]

yeah. The question remains as to how much of a clock improvement can AMD get with Pinnacle Ridge. If AMD can hit 4 Ghz base and 4.5 Ghz max turbo with avg max OC of 4.6-4.7 Ghz they will have a very competitive product. So it all boils down to how good the 12LP process is and has AMD made improvements to the Zen physical design to extract any more clock speed headroom.

Slides indicate a 10% boost. With no uarch changes should mean a boost to Fmax. I would expect stock clocks of 4.0GHz w/ 4.4GHz boost. This will make for very attractive CPUs, especially for Ryzen 7. We'll just have to wait and see for final overclocking limits.

Lastly - wow, Pinnacle Ridge must actually be in 'risk' production now - AMD and GF are sandbagging in their slides! As had been said, this is a new AMD and a new GloFo!

 

[scannall]

Slides indicate a 10% boost. With no uarch changes should mean a boost to Fmax. I would expect stock clocks of 4.0GHz w/ 4.4GHz boost. This will make for very attractive CPUs, especially for Ryzen 7. We'll just have to wait and see for final overclocking limits.

Lastly - wow, Pinnacle Ridge must actually be in 'risk' production now - AMD and GF are sandbagging in their slides! As had been said, this is a new AMD and a new GloFo!

They sandbagged Ryzen performance as well, saying it would be %40 faster than the previous construction core. Turned out to be %52. Glad to see them being aggressive. They really need to be at this point.

 

[Arachnotronic]

It's pretty interesting how AMD's and GloFo's execution seemingly continue to improve as Intel's degrades. Will be interesting to see if GloFo/AMD keep it up at 7nm.

 

[Phynaz]

Slides indicate a 10% boost. With no uarch changes should mean a boost to Fmax. I would expect stock clocks of 4.0GHz w/ 4.4GHz boost.

Careful, slide says up to 10% performance improvement. It does not say what the measurement of performance is.

Extrapolating a GloFo slide to what "should" happen to the operating frequency of an AMD CPU is a pretty risky jump. Even if the stated performance improvement is Fmax of a transistor, that does not necessarily translate into into the Fmax of a CPU that is made up of a few billion of those transistors.

Basically, a slide from GloFo doesn't provide enough information to predict the performance of an unannounced CPU.

 

[Ajay]

Careful, slide says up to 10% performance improvement. It does not say what the measurement of performance is.

Extrapolating a GloFo slide to what "should" happen to the operating frequency of an AMD CPU is a pretty risky jump. Even if the stated performance improvement is Fmax of a transistor, that does not necessarily translate into into the Fmax of a CPU that is made up of a few billion of those transistors.

Basically, a slide from GloFo doesn't provide enough information to predict the performance of an unannounced CPU.

True, there could be small changes that have enhanced Ryzen performance beyond just the clock increases. I'm really expecting a boost in max memory overclocks, but we'll see - AMD may have focused more on small changes to improve EPYC performance.

 

[Arachnotronic]

True, there could be small changes that have enhanced Ryzen performance beyond just the clock increases. I'm really expecting a boost in max memory overclocks, but we'll see - AMD may have focused more on small changes to improve EPYC performance.

12LP won't be used for EPYC. Client only. Next EPYC is Rome on 7LP.

 

[Ajay]

12LP won't be used for EPYC. Client only. Next EPYC is Rome on 7LP.

Yeah, saw that but wasn't sure - thanks! Makes sense, why validate another server CPU this close to 7nm.

Edit to spell correctly - dammich!

 

[Arachnotronic]

Yeah, saw that but wasn't sure - thanks! Makes sense, why valid another server CPU this close to 7nm.

Yep. AMD will also get much more revenue from Pinnacle Ridge than it reasonably could from a Pinnacle Ridge-derived EPYC.

 

[CatMerc]

Careful, slide says up to 10% performance improvement. It does not say what the measurement of performance is.

Extrapolating a GloFo slide to what "should" happen to the operating frequency of an AMD CPU is a pretty risky jump. Even if the stated performance improvement is Fmax of a transistor, that does not necessarily translate into into the Fmax of a CPU that is made up of a few billion of those transistors.

Basically, a slide from GloFo doesn't provide enough information to predict the performance of an unannounced CPU.

Very important note!!! ^^^

It's 10% performance at ISO power, it doesn't say anything about Fmax. Which is both a good thing and a bad thing. A bad thing because it means we don't know what Fmax will look like. Good because there's thew possibility of Fmax being higher than 10%, even if power consumption goes haywire.