Intel "Willow Cove" Backported to 14nm is "Cypress Cove"?

[btarunr]

Intel's 11th generation Core "Rocket Lake-S" desktop processor is fascinating as it introduces Intel's first CPU core IPC uptick in about half a decade. Until now, it was rumored that "Rocket Lake-S" features a back-port of Intel's "Willow Cove" CPU cores to the 14 nm silicon fabrication process. It turns out that Intel doesn't want to call these cores "Willow Cove," which make their debut with the 10 nm+ "Tiger Lake" mobile processors later this Summer. Enter "Cypress Cove." A Moore's Law is Dead video presentation sheds light on this mysterious new codename.

Apparently, "Cypress Cove" is the codename Intel is using to refer to the CPU cores Intel is building with its latest CPU core IP on older 14 nm process. Owing to the process, the IPC of these cores may be different from the "Willow Cove" cores on "Tiger Lake," and to avoid confusion, Intel possibly choosing to give it a different internal codename. In other words, Moore's Law is Dead believes that "Cypress Cove" may not offer the alleged 25% IPC gains over "Skylake" that you could expect instead from "Willow Cove" cores in "Tiger Lake."



The maximum core count of "Rocket Lake-S" is swinging between 8 and 10, although more sources lean toward 8 than 10. Taking advantage of the increase power budget of the desktop platform, "Cypress Cove" cores could be clocked a lot higher than "Willow Cove" cores on mobile chips such as "Tiger Lake." Next up, while "Rocket Lake-S" is expected to feature Gen12 Xe graphics, it could have a lower execution unit (EU) count than "Tiger Lake," which has 96 of them. Sources of Moore's Law is Dead reinforce the theory from last week of "Rocket Lake-S" being a multi-chip module of CPU cores sitting on a 14 nm die, and the iGPU along with other uncore components on a separate 10 nm die.

Sources tell Moore's Law is Dead that "Rocket Lake-S" is heading toward either a Q4-2020 or Q1-2021 launch, although they lean towards the latter. Find other fascinating insights in the source link below.

View at TechPowerUp Main Site

 

[Vayra86]

"To avoid confusion, Intel is giving it a new codename"

euhm

 

[R0H1T]

So you thought Intel was drowning you in lakes

 

[micropage7]

"To avoid confusion, Intel is giving it a new codename"

euhm

every little update is a new name, sometimes i think that Intel try to poison our mind that new name is a new architecture with 20% improvement actually not like that in real world

 

[Space Lynx]

I just wish they would focus more on IPC, speed improvements, and leave everything at 8core 16thread

 

[ppn]

The IPC is there, but until DDR5-4800 may not manifest itself very well. Everything is stuck now. Not constrained by 14nm, but other things. It will all come together nicely by 2024.

 

[cucker tarlson]

MLID with another fake MS paint slide
he should really trade that amd 50th anniversariy shirt for some basic image creation software

 

[dicktracy]

turns out the only one that will beat grandpa Skylake in gaming is... another chip from Intel! Can’t wait for Rocket Lake. We are going to space with Rocketlake while AMD is still figuring out how to defeat a 90 year old in a wheelchair!

 

[JAKra]

Maybe the alleged 25% IPC gains over "Skylake" are cumulative.
Willow Cove: +9%
Cypress Cove: +6%
Dead Cow: +10

 

[cucker tarlson]

turns out the only one that will beat grandpa Skylake in gaming is... another chip from Intel!

this is just sad really.
skylake is 5 years old and made on 14nm.
we should really have cpus that run circles around it in 2020.

 

[Bwaze]

In other words, Moore's Law is Dead believes that "Cypress Cove" may not offer the alleged 25% IPC gains over "Skylake" that you could expect instead from "Willow Cove" cores in "Tiger Lake."

II imagine this is because they had to cut whole areas of cache and maybe even simplify architecture, cut pipelines to squeeze it into 14nm process - it's not nearly enough to limit it to 8 cores instead of 10.

Planned transistor density of 10nm process for wich "Willow Cove" architecture was designed is vastly greater than in ancient 14 nm process:

You can't just "print it bigger"...

 

[Gungar]

II imagine this is because they had to cut whole areas of cache and maybe even simplify architecture, cut pipelines to squeeze it into 14nm process - it's not nearly enough to limit it to 8 cores instead of 10.

Planned transistor density of 10nm process for wich "Willow Cove" architecture was designed is vastly greater than in ancient 14 nm process:

You can't just "print it bigger"...

Yes you can just print it bigger with some ajustements obviously.

 

[Caring1]

turns out the only one that will beat grandpa Skylake in gaming is... another chip from Intel! Can’t wait for Rocket Lake. We are going to space with Rocketlake while AMD is still figuring out how to defeat a 90 year old in a wheelchair!

 

[cucker tarlson]

Skylake is the Ridge Forrester od cpus.
Put some makeup and a wig on it and its ready for another five years.

 

[Tomgang]

Are you kidding me, 14 nm on another gen again.

Intel is never moving on to 10 nm for desktop cpu cores. That is at least how it feels. If this continues like this and depending on my economy at the end of this year, if I end up upgrading. I think AMD ZEN 3 sounds like a better option now.

 

[Bwaze]

Yes you can just print it bigger with some ajustements obviously.

Intel bragged in 2018 that 10nm node is 2.7x denser than 14nm. And I believe they didn't plan to use all that density just to make smaller chips, even Ice Lake "Sunny Cove" architecture increased transistor count - larger caches, wider architecture, AVX512... And now we have another increase in transistor count - I think printing it 2.7x bigger can quickly get you in a realm of several thousands of dollars for each monolithic die!

 

[Vayra86]

Intel is never moving on to 10 nm for desktop cpu cores

That is exactly what I have been convinced of since 2018

So far so good

Maybe the alleged 25% IPC gains over "Skylake" are cumulative.
Willow Cove: +9%
Cypress Cove: +6%
Dead Cow: +10

Haha I see what you did there

 

[stimpy88]

II imagine this is because they had to cut whole areas of cache and maybe even simplify architecture, cut pipelines to squeeze it into 14nm process - it's not nearly enough to limit it to 8 cores instead of 10.

Planned transistor density of 10nm process for wich "Willow Cove" architecture was designed is vastly greater than in ancient 14 nm process:

You can't just "print it bigger"...

The size of the die is really not an issue here. The issue is that Intel want $$$ from each wafer they produce, and are not willing to sacrifice their die per wafer ratio, thus their profit margins. It's ok that their product is crap, fanbois still buy them.

 

[my_name_is_earl]

Intel seem to be sitting comfortably on an ancient tech. That's how you get left in the dirt. Just like Blackberry vs Apple.

 

[Tomgang]

That is exactly what I have been convinced of since 2018

So far so good



Haha I see what you did there

It's literally a joke now. They been on 14 NM since 2014/2015 for laptops and desktop and this seems to continue in to 2021 as well now. I am not impressed by Intel any more. I have been a Intel supporter for years, but lately I am beginning to have more faith in AMD hornestly.

 

[chris.london]

Who wants to bet that Intel will market Rocket Lake as 10nm because the uncore will be manufactured on 10nm?

 

[fynxer]

14nm, the never ending saga...

 

[Bruno Vieira]

This 8-core 14nm die will be HUGE, and they will fight zen3 with true 8-cores as well. It is going to be interesting.

 

[Verpal]

Who wants to bet that Intel will market Rocket Lake as 10nm because the uncore will be manufactured on 10nm?

If they manufacture with 10nm core and 14nm uncore, advertising it as 10nm would not be extremely problematic, but if we are talking about 10nm uncore only, class action lawsuit awaits.

 

[yeeeeman]

This is not good news, but at least it is some news.
The only lesson we can learn from this is that messing up just one node in the fabrication process can be an extremely hard problem to fix.
I would say that part of the reason why Intel took so long to admit they have a problem and find a solution is the internal bureaucracy and bad culture. A big mammoth company is much slower to adapt to problems and that was very obvious.
Anyway, if Rocket Lake comes out and has better (and I mean, clearly better) gaming performance with reasonable power consumption and similar IPC to Zen 3, I don't think it will be a tragedy for Intel. But if not, they might be in big trouble on the desktop segment in the next year or so.

I wonder what Intel would have done if they didn't have this good Skylake core to keep them in the race during all these years...