AMD's new Zen 6 chips will have a low-power core built from five different generations of Zen

[Skye Jacobs]

What we know so far: AMD is taking a different approach with its upcoming Zen 6 processors, introducing a new type of low-power core built from pieces of several earlier Zen designs rather than a single architecture. The change surfaced in a recent Linux kernel patch, which shows how the system identifies and manages these cores. The patch makes clear that the low-power cores are distinct enough to warrant their own classification, rather than being grouped with AMD's existing compact "C" cores.

Patch notes that the system needs a way to properly identify these low-power cores in user space, since they currently show up as "unknown" in the CPU topology.

It also specifies that on supported AMD and Hygon chips, these cores should scale performance using amd_get_highest_perf() instead of the fixed CPPC_HIGHEST_PERF_PERFORMANCE limit, aligning them with how existing efficiency cores are managed.

What stands out is how these cores are built. Analysis from leaker InstLatX64 points to a hybrid design that combines multiple generations of Zen. The reported setup includes a Zen 6 instruction set, Zen 5 microarchitecture, Zen 4 floating-point unit, Zen 3 L2 cache, and Zen 2 L3 cache.

– InstLatX64 (@InstLatX64) July 29, 2026

Instead of designing a low-power core from scratch, AMD appears to be reusing and combining existing components. That mix-and-match approach suggests the company is trying to improve efficiency while keeping development time and complexity in check.

These low-power cores would sit alongside AMD's current Performance and Efficiency cores, creating a three-tier setup. Performance cores handle high-clock workloads, while Efficiency cores are smaller and tuned for lower power use. The new low-power cores go a step further, targeting very light workloads where minimizing power draw is the priority. In that sense, they're closer in purpose to Intel's low-power efficiency cores, though AMD's design is structured differently.

The potential benefit is clearer in mobile devices. Laptops and gaming handhelds could offload lighter tasks to these low-power cores, reducing overall power consumption. That could help extend battery life and let systems run certain processes at just a few watts. How much of a difference that makes will depend heavily on how well software can manage and assign workloads across the different core types.

There are still gaps in what's known. AMD hasn't shared performance or efficiency data for these cores, and it's not yet clear how well this blended architecture will scale under real workloads. The design is unconventional, and its success will likely hinge on how smoothly the hardware and software layers work together.

This development comes as AMD is rolling out its Zen 6 server chips, with desktop versions expected later this year. But the broader market may complicate things. PC shipments have been trending down, and higher memory prices, driven in part by AI demand, are pushing up system costs.

That matters more in laptops and handhelds, where upgrading the CPU usually means replacing the entire device. Even if AMD delivers better efficiency, adoption could be limited if buyers hold off on new hardware.

Still, the idea of building a core from a mix of proven parts marks a notable change in direction. Rather than stretching one architecture across every use case, AMD is carving out more specialized roles for different core types. If it works, it could give the company more flexibility in how it designs chips for power-sensitive devices.

Permalink to story:

AMD's new Zen 6 chips will have a low-power core built from five different generations of Zen

 

[Beerfloat]

AMD used to say Intel's hybrid Power/Efficiency cores added unnecessary scheduling and software complexity, but we all learn as we go along.

 

[Inthenstus]

If AMD has loser cores I’m just going to stop buying CPUs lol. Horrible play by team Red.

 

[Anton Longshot]

Interesting developments.

Looking forward to seeing how the final product will do.

As an aside: for a while now I've been interested in building a desktop with laptop components as my secondary machines don't really need a lot of performance; only my game box uses a bit of that.
Still, laptop components that can be used in a desktop aren't easy to find.
I'm hoping this will change some time soon-ish.

 

[Linux Mint Xfce]

It’s alive! It’s alive!!

 

[cfenton]

I'm surprised there's anything from Zen 2 that is still more efficient than Zen 6.

 

[GeoffreyA]

I'm surprised there's anything from Zen 2 that is still more efficient than Zen 6.

Almost always trade-offs going from one generation to the next. Zen 2's L3 cache had slightly lower latency than Zen 3's.

 

[yRaz]

I guess this is kinda cool. I dont see it as really applying to me and I would put money on the fact that I use my laptop away from a plug more than anyone else on the forum.

I have often wondered why we insist on making faster cores for everything. I will also kick the first person to bring up ARM in the nads so hard they'll fly out yoyr mouth at hypersonic speed. The 7735hs in my ThinkPad has never left me wanting in the field but its also overkill for looking at schematics and arguing with HR

 

[hwertz]

I'll just note, in a thermally constrained setup (vendors insist on making notebooks thin rather than just a bit thicker for cooling), these will probably help your heavy workloads run at higher boost clock for longer since the light work will generate that bit less heat.

 

[Axeia]

*deletes entire post*
Nevermind what I initially said. It sounds interesting although I don't quite understand what it entails. First I thought it was going to be multiple different cores (like Intel) on the same die.
But it's a whole puzzle of (presumably) the best parts of various architectures per core?
Think I'll just await more news.

 

[Theinsanegamer]

I guess this is kinda cool. I dont see it as really applying to me and I would put money on the fact that I use my laptop away from a plug more than anyone else on the forum.

I have often wondered why we insist on making faster cores for everything. I will also kick the first person to bring up ARM in the nads so hard they'll fly out yoyr mouth at hypersonic speed. The 7735hs in my ThinkPad has never left me wanting in the field but its also overkill for looking at schematics and arguing with HR

Because you can do more work with a faster core. Duh? I mean, why do we make anything faster? Why would we stop progress at some arbitrary point in time?

Also, you know, race to idle? People always forget that exists for some reason.

If AMD has loser cores I’m just going to stop buying CPUs lol. Horrible play by team Red.

Oh no, he cant buy only big cores, the low power core is going to make him angwy uwu X(

Buddy, its a CPU core. Disable it if you want to burn power watching youtube.

 

[yRaz]

Because you can do more work with a faster core. Duh? I mean, why do we make anything faster? Why would we stop progress at some arbitrary point in time?

Also, you know, race to idle? People always forget that exists for some reason.
Oh no, he cant buy only big cores, the low power core is going to make him angwy uwu X(

Buddy, its a CPU core. Disable it if you want to burn power watching youtube.

No, it is that the work I've been doing for decades has not required more compute aside from the fact that M$ insists I need to have a better map of my colon

 

[Mr Majestyk]

AMD used to say Intel's hybrid Power/Efficiency cores added unnecessary scheduling and software complexity, but we all learn as we go along.

They said that because e-cores and P cores were fundamentally totally different cores. C cores are identical to regular cores other than cache sizes and to the OS they are identical which was not the case for e and p cores.

 

[HardReset]

They said that because e-cores and P cores were fundamentally totally different cores. C cores are identical to regular cores other than cache sizes and to the OS they are identical which was not the case for e and p cores.

To be more precise, normal and c cores from same generation are different. However outside caches and perhaps heavy AVX, they have same IPC and software sees them as identical.

These new cores probably have very small clock speed Zen6 "main" cores mainly for integer performance and just to save work not so important parts are recycled. Sounds pretty clever on paper.

 

[Inthenstus]

Because you can do more work with a faster core. Duh? I mean, why do we make anything faster? Why would we stop progress at some arbitrary point in time?

Also, you know, race to idle? People always forget that exists for some reason.
Oh no, he cant buy only big cores, the low power core is going to make him angwy uwu X(

Buddy, its a CPU core. Disable it if you want to burn power watching youtube.

Oh yeah like 99% of OEMs let you disable cores now? Since when? To this day many applications are still sluggish because they end up on these low power cores because your computer determines that you want you apps to load slowly. No thanks.

 

[acrossthestars]

ISA: #Zen6
uArch: #Zen5 [2]
FPU: #Zen4 (256b)
L2: #Zen3 (512KB) [1]
L3: #Zen2/#Mendocino (1MB/core)

1. Is there actually enough Zen 4, 3 and 2 supply to create this "CollaguePU"? These could be binned parts, but I thought AMD was already using binned parts. Zen 2 is from before Covid. How are fabricators going to handle this multi-generational order? Each generation is built on a different process node? If they run out of some parts and not others, is TSMC going refire the 7nm furnance and intentionally build "pre-outdated parts" to make up for the now-exhausted supply, or are they going to simply label some parts as Zen 3 just to avoid the complication.

2. How is this going to work, as a product? I know that AM4 was granted compatibility with Zen 1 through 3 and AM5 is Zen 4 and 5, but every chip was monolithic. A Ryzen 5600 G CPU was completely Zen 3, it wasn't 50% Zen 3, 25% Zen 2 and 25% Zen 1. How are the different generations' architectures going to talk to each other? Is there even precedent for a system like this? big.Little architecture usually means different families of the same chipset talking to each other, not different generations.

This feels like AMD wanted to do Low-Power and High-Performance cores too, but that would have been too much like Intel and ARM. AMD is special. They want to be different(TM) and what's more different than a franken-CPU, comprised of 4 different generations of Ryzen processors, all stitched together?

3. Doesn't using older generations of Ryzen architecture for certain tasks sort of imply that AMD's architectural synthesis is getting worse with time, not better? Like, if each new generation was improving across the board, why are we using older tech for better power efficiency? Wouldn't the usage of 7nm architecture for L3 cache negate the power envelope difference of Zen 3 for L2 cache or am I missing something?

This whole endeavor sounds like what Apple did with the Macbook Neo, except that they quickly ran out of binned A18 Pro chips, because Tim Apple is really good at understanding supply chain logistics, but real sh*t at understanding the value proposition of a device that would have cost $1,100 10 years ago, for $600. So, they had to send in another order, meaning that fabricators were intentionally building outdated parts, when they could be making more money fabricating the newer stuff. It also just comes across as implementing cost-cutting measures in the dumbest, penny-pinching method possible. At least Apple was using entire binned A18 parts, they didn't somehow cut some A17 Pro into the mix, for good measure.

AMD should go back to the drawing board and start from scratch, if a monolithic Zen 6 with low- and high-power cores is less capable than one built piecemeal from older parts.

 

[GeoffreyA]

1. Is there actually enough Zen 4, 3 and 2 supply to create this "CollaguePU"? These could be binned parts, but I thought AMD was already using binned parts. Zen 2 is from before Covid. How are fabricators going to handle this multi-generational order? Each generation is built on a different process node? If they run out of some parts and not others, is TSMC going refire the 7nm furnance and intentionally build "pre-outdated parts" to make up for the now-exhausted supply, or are they going to simply label some parts as Zen 3 just to avoid the complication.

2. How is this going to work, as a product? I know that AM4 was granted compatibility with Zen 1 through 3 and AM5 is Zen 4 and 5, but every chip was monolithic. A Ryzen 5600 G CPU was completely Zen 3, it wasn't 50% Zen 3, 25% Zen 2 and 25% Zen 1. How are the different generations' architectures going to talk to each other? Is there even precedent for a system like this? big.Little architecture usually means different families of the same chipset talking to each other, not different generations.

This feels like AMD wanted to do Low-Power and High-Performance cores too, but that would have been too much like Intel and ARM. AMD is special. They want to be different(TM) and what's more different than a franken-CPU, comprised of 4 different generations of Ryzen processors, all stitched together?

3. Doesn't using older generations of Ryzen architecture for certain tasks sort of imply that AMD's architectural synthesis is getting worse with time, not better? Like, if each new generation was improving across the board, why are we using older tech for better power efficiency? Wouldn't the usage of 7nm architecture for L3 cache negate the power envelope difference of Zen 3 for L2 cache or am I missing something?

This whole endeavor sounds like what Apple did with the Macbook Neo, except that they quickly ran out of binned A18 Pro chips, because Tim Apple is really good at understanding supply chain logistics, but real sh*t at understanding the value proposition of a device that would have cost $1,100 10 years ago, for $600. So, they had to send in another order, meaning that fabricators were intentionally building outdated parts, when they could be making more money fabricating the newer stuff. It also just comes across as implementing cost-cutting measures in the dumbest, penny-pinching method possible. At least Apple was using entire binned A18 parts, they didn't somehow cut some A17 Pro into the mix, for good measure.

AMD should go back to the drawing board and start from scratch, if a monolithic Zen 6 with low- and high-power cores is less capable than one built piecemeal from older parts.

I don't think it's this. Each chiplet is monolithic and it is only at the source or RTL level that parts can be mixed and matched. What's happening, I would guess, is that they are mixing design pieces from several architectures, with a view towards power efficiency in those cores.

Though each architecture was a unity, it was always a revision of the previous; it is fair to say, linguistically, that Zen 3 used pieces from Zen 2. For example, the micro-op cache was unchanged.

Using older pieces doesn't mean the newer architectures are deteriorating. With every advance, there's a trade-off regarding performance, power, and area. So, it makes sense, if one were optimising for a certain metric, to use a design motif consistent with that end. Zen 4 doubled the L2 cache; therefore, if optimising for area, using Zen 3's smaller L2 cache would be the choice.

 

[Endymio]

I don't think it's this. Each chiplet is monolithic and it is only at the source or RTL level that parts can be mixed and matched. What's happening, I would guess, is that they are mixing design pieces from several architectures, with a view towards power efficiency in those cores.

Exactly so. AMD isn't literally cutting out bits of actual chips and pasting them together; the "collage" is of design elements, not physical ones.

 

[HardReset]

I don't think it's this. Each chiplet is monolithic and it is only at the source or RTL level that parts can be mixed and matched. What's happening, I would guess, is that they are mixing design pieces from several architectures, with a view towards power efficiency in those cores.

Though each architecture was a unity, it was always a revision of the previous; it is fair to say, linguistically, that Zen 3 used pieces from Zen 2. For example, the micro-op cache was unchanged.

Using older pieces doesn't mean the newer architectures are deteriorating. With every advance, there's a trade-off regarding performance, power, and area. So, it makes sense, if one were optimising for a certain metric, to use a design motif consistent with that end. Zen 4 doubled the L2 cache; therefore, if optimising for area, using Zen 3's smaller L2 cache would be the choice.

Yeah. Major part of architecture is very probably Zen5 c-cores (scheduling, integer units etc). No idea why Zen6 c-cores are not used. Probably there was too much work with Zen6 based server CPUs or something. They however support Zen6 ISA so that CPU should have longer lifecycle (it can be considered to be using "latest architecture" even when it does not). FPU calculations are rare on low power chips so no reason to put effort on there. AVX is probably "supported" but who actually will use that? Therefore Zen4 FPU makes lot of sense.

For caches, AMD decided to take most suitable cache configuration and structure without having to design anything new.

All in all this seems to be "make something using as much older stuff as possible" -solution. Very low effort but it may still work well. Even if it does not, it's still low effort.

 

[GeoffreyA]

All in all this seems to be "make something using as much older stuff as possible" -solution. Very low effort but it may still work well. Even if it does not, it's still low effort.

Also goes to show how well those older pieces were engineered.

 

[Watzupken]

Its funny when Intel is looking to unify the 2 types of cores to mimic AMD, AMD is mimicking Intel by splitting the core types. Each solution comes with their own pros and cons. I may be wrong, but Intel's current solution is more heavily dependent on software.

https://www.techpowerup.com/346645/...no-more-performance-and-efficiency-core-split

 

[HardReset]

Its funny when Intel is looking to unify the 2 types of cores to mimic AMD, AMD is mimicking Intel by splitting the core types. Each solution comes with their own pros and cons. I may be wrong, but Intel's current solution is more heavily dependent on software.

https://www.techpowerup.com/346645/...no-more-performance-and-efficiency-core-split

Not exactly. AMD is not splitting core types since both AMD "normal" core and AMD c-core share exactly same IPC and same architecture from software POV (except caches). Basically AMD have same cores, some cores have lower clock speed and that's it. Intel had two totally different core architectures and different IPC on same package. That causes many problems.

 

[Aranarth]

1. Is there actually enough Zen 4, 3 and 2 supply to create this "CollaguePU"?

2. How is this going to work, as a product?


3. Doesn't using older generations of Ryzen architecture for certain tasks sort of imply that AMD's architectural synthesis is getting worse with time, not better?

You seem to not understand how CPUs are created.

This is still all new silicon that will work in the am5 socket.
This is analogus to building a new car but using the technology from previous years.

Say I used a body design from c4 corvette, wheels from c7, transmission and engine from a c8, differential from a c6. seats from a c7 etc. all brand new on a sales floor?

In this case it is parts from previous generations of Zen but still built on new silicon so it can work side by other cores. They used used the lowest power parts from previous generations to make the lowest power cpu avaiable. it is STILL ryzen. it is not a completely different core from high power cores which is what intel did to create efficiency and high power cores.

 

[JamesBlond]

If AMD keep pushing cores not new tech they will never gain, they have to do something that is bigger than the current move...like 1 core on the CPU that is responsible for compressing data at lighting speed, 1 that is tasked at controlling special functions, the same thing faster is falling behind, new...that is what the world needs

 

[HardReset]

If AMD keep pushing cores not new tech they will never gain, they have to do something that is bigger than the current move...like 1 core on the CPU that is responsible for compressing data at lighting speed, 1 that is tasked at controlling special functions, the same thing faster is falling behind, new...that is what the world needs

Truth is, AMD has very low interest for notebook CPUs/APUs because people will buy Chipzilla anyway. Therefore AMD don't bother much and makes something new without putting too much effort.