AMD Zen 4C: The return of Athlon and other lower end SKUs? (a discussion)

[TimedPing]

As many of you have probably know, recently AMD release their version of "Efficiency cores" called Zen 4C for latest SP5 based EPYC bergamo CPU, even though it's not quite an efficiency cores like what Intel did with their Gracemont E-Cores. Zen 4C is basically a slightly cut down Zen 4 cores, with lower clock speed and halves the L3 Caches, which enable AMD pack more cores per CCD die (Details on spoilers below):

Spoiler

Source: https://www.techpowerup.com/310057/amd-zen-4c-not-an-e-core-35-smaller-than-zen-4-but-with-identical-ipc

Based on this, can we optimistically expect AMD to release the consumer version of this new cores into lower end SKUs like Desktop/Mobile Athlon CPU/APU and some embedded SKUs as well like Intel Alder Lake N series? We know AMD doesn't gonna waste those slightly defective Zen 4C cores, just take a look with recently launched Ryzen 5 5600X3D and rumored upcoming late-launched low-end SKUs on AM4 platform? Let me know what you guys think on this interesting take!

[Jurrunio]

Mobile platform first if they are coming to consumers, not so much desktops imo. Really low end desktops are just super rare these days, people who bought those just choose laptops instead because it doesn't take much for the average person to buy the idea of portability.

[porina]

On reusing "defective" cores, feels unlikely unless they do a hybrid desktop chiplet CPU with up to 8C+16c. This wouldn't work exactly like Intel's hybrid CPUs though so it may scale differently.

 

On using the smaller cores in lower offerings, there I'm also not sure. Part of the design change required to shrink them limits their clock potential. I'm not sure that would be acceptable to only use the c cores where APUs are deployed for value desktops and general laptops. Existing APUs already have half L3 cache compared to the equivalent chiplet desktop equivalents. They don't have the clock nerf though. c core only APUs could exist for really low end offerings, we're looking at dedicated embedded systems, maybe Chromebook type class.

 

It is more of a possibility if they follow a more Intel style integrated CPU, where there is a mix of C and c cores. Currently we have up to 8C in APUs. For the same area they could move to 6C+4c for example. This should increase MT performance while still having sufficient big cores for most uses. APUs being monolithic means they'll have shared L3, which is not possible on chiplet implementation.

[YoungBlade]

AMD desktop APUs are just laptop chips soldered to a desktop CPU substrate. This means that we would likely see AMD laptop chips with Zen 4C cores before desktop ones. There were rumors of AMD testing a 2+4 big.LITTLE configuration back around March, apparently codenamed Phoenix 2, so it's possible we'll see a low-end consumer APU using Zen 4C cores at some point.

 

That said, these tests were with monolithic dies, so the defective yields from EPYC are not the source of them. IMO, those are unlikely to be used for consumer chips. Instead, I imagine AMD will try to use as many as possible for cut down EPYC CPUs, as those have a way higher margin and have a market for high core counts over single core performance.

 

But we'll see what happens. Perhaps there will be some Zen 4C low-end desktop chips, like a 4c/4t Athlon using a defective Zen 4C dies with half the cores disabled. It's not totally out of the question, but it doesn't seem likely to me.

[starsmine]

3 hours ago, YoungBlade said:

But we'll see what happens. Perhaps there will be some Zen 4C low-end desktop chips, like a 4c/4t Athlon using a defective Zen 4C dies with half the cores disabled. It's not totally out of the question, but it doesn't seem likely to me.

one zen 4c chiplet is 16 cores. 
I doubt there are many that chiplet SO defective that running them in 4c mode makes for a plausible sku.

[YoungBlade]

21 minutes ago, starsmine said:

one zen 4c chiplet is 16 cores. 
I doubt there are many that chiplet SO defective that running them in 4c mode makes for a plausible sku.

I apparently misremembered the details. Zen 4C uses two 8 core CCXs per CCD, like Zen 2, whereas I thought each CCX was a totally independent CCD.

 

In that case, a 4c SKU from defective Bergamo dies is even more unlikely than the level of unlikely I already posited.

[TimedPing]

3 hours ago, starsmine said:

one zen 4c chiplet is 16 cores. 
I doubt there are many that chiplet SO defective that running them in 4c mode makes for a plausible sku.

Well, R5 5600X3D exist, so I have little doubt AMD will not use these defective chiplet into new SKUs