AMD Zen 6 desktop CPUs may deliver 24 cores, 7 GHz targets, and AM5 support

Daniel Sims

Posts: 2,479   +75
Staff
Rumor mill: It's long been rumored that AMD's next desktop CPU architecture, Zen 6, could finally break from the eight-cores-per-chiplet formula the company has stuck with for years. New details now sketch out multiple core configurations, including top-end parts scaling to 20 and even 24 cores.

According to well-known leaker HXL, Zen 6 will span four single-CCD processor variants alongside three higher-end dual-CCD designs. While the exact Ryzen branding for each chip hasn't been disclosed, the rumored Ryzen 10000 series lineup appears to back earlier whispers that AMD's next flagship desktop CPU will push core counts into unfamiliar territory.

On the lower and midrange side, HXL lists four single-chiplet CPUs with six, eight, 10, and 12 cores. The 12-core model would logically slot in as a Ryzen 7, but the presence of a six-core option also hints that AMD could resurrect Ryzen 3 branding at the bottom of the Zen 6 stack.

At the top end, three dual-CCD CPUs reportedly scale from 16 cores (8+8) to 20 cores (10+10) and finally 24 cores (12+12). AMD has capped each CCD at eight cores since Zen 2, but leaks dating back to last year suggested that Zen 6 would raise that limit to 12, at least for the highest-tier parts.

Clock speeds could also climb to new highs. Zen 4 topped out at 5.7 GHz, but Zen 6 is said to have hit 6.5 GHz in internal testing, with AMD allegedly aiming for the symbolic 7 GHz milestone.

Cache details are less clear. Some reports point to 48 MB of L3 cache per CCD, which would give the 24-core flagship a total of 96 MB. Others suggest a far more aggressive 144 MB per CCD, surpassing the 128 MB used on Zen 5 X3D dies and potentially yielding a staggering 288 MB of L3 cache on a top-end model.

Intel's upcoming Nova Lake desktop CPUs are rumored to pursue similar ideas as part of the company's first serious counter to AMD's 3D V-Cache approach.

AMD has already confirmed that Zen 6 will arrive later this year and move to a 2nm manufacturing process, likely TSMC's N2P or N2X nodes. Nova Lake is also expected in late 2026 on a 2nm process from the same foundry.

Unlike Intel's next platform – which will require a new motherboard socket – AMD has said Zen 6 will remain compatible with existing AM5 boards. Beyond desktops, Zen 6 is expected to show up in notebooks and may even power the next Xbox sometime in 2027.

Permalink to story:

 
I have one more aged PC to rebuild - an Ivy Bridge box. I'm looking forward to replacing it with Zen 6, and I'm quite pleased Zen 6 will support AM5. I want to upgrade one proc with a Zen 6 part. Hopefully, benchmarks will convince me it is worth the cost.
 
I have one more aged PC to rebuild - an Ivy Bridge box. I'm looking forward to replacing it with Zen 6, and I'm quite pleased Zen 6 will support AM5. I want to upgrade one proc with a Zen 6 part. Hopefully, benchmarks will convince me it is worth the cost.
I'm sure the CPUs will be reasonably priced but the ram you need for it won't be. I'm sure selling yourself on the CPU will be easy.
 
In the era of AI, memory is the new logic.

Consequently, CPUs require significant amounts of L4 cache, measured in gigabytes. In the future, as manufacturing yields also reach their peak, the distinguishing feature of CPUs we purchase may well be the size of their L4 cache. The number of cores and total instructions per cycle (IPC) will likely become standard. For instance, we might buy a Zen 7 processor featuring 16 cores and L4 HBM cache options of 128, 256, or 512 gigabytes.
Intel-Sapphire-Rapids-696x488.jpg
 
Last edited:
My 7700x in my main rig is still doing just fine. I'm sure when they're replaced by AM6 in a couple years I'll upgrade it to whatever the best 10000x3d chip is but until then I'm good.
 
Dear Mr Daniel Sims,
It's not either/or, it's both to exist: 96 MB (without 3DVCache) and 288 MB (with 3DVCache).
 
... CPUs require significant amounts of L4 cache, measured in gigabytes. In the future, as manufacturing yields also reach their peak, the distinguishing feature of CPUs we purchase may well be the size of their L4 cache. The number of cores and total instructions per cycle (IPC) will likely become standard. For instance, we might buy a Zen 7 processor featuring 16 cores and L4 HBM cache options of 128, 256, or 512 gigabytes.
HBM is the main memory, not cache! SRAM is used for cache because of its low latency as well as native speed. DRAMs, like HBM, require a lot of SRAM around them to go fast. Same for Flash. But SRAM is bulky on the silicon - it's built from the same transistors as the fast logic is. It takes up a lot of space.

IBM Z series is showing the way forward. A few year back they did away with both L3 and L4. Virtualising both into much much larger blocks of physical L2.
 
The line up that I see as natural would be:
10500X 6 cores
10600X 8 cores
10700X 10 cores
10800X 12 cores
10900X 16 cores
10920X 20 cores
10950X 24 cores

And eventually they should offer X3D versions of the entire range.

They can finally truly differentiate 700X and 800X parts by core count, not just v-cache. I would be more than happy with a plain old 12 core 10800X, rather than pay a huge premium for v-cache. This would still thrash a 9900X in gaming.
 
Going to look forward to replacing my 9800X3D with a 10800X3D.

And thankfully I bought ram in april 2025 paid $300 for my 64GB kit which now cost $1500.
 
It means that the server side (Epyc) will likely start having up to 192 cores and 384 usable threads.

It's insane the push AMD has been doing.
 
A 24-core desktop chip on AM5 feels absurd in a good way. We’re basically at the point where consumer CPUs are brushing up against old HEDT territory, but without the platform tax or quad-channel memory baggage.
 
It means that the server side (Epyc) will likely start having up to 192 cores and 384 usable threads.

It's insane the push AMD has been doing.
That's customer demand, not push. Servers actually use those cores.

A 24-core desktop chip on AM5 feels absurd in a good way. We’re basically at the point where consumer CPUs are brushing up against old HEDT territory, but without the platform tax or quad-channel memory baggage.
You do know Intel has been pushing desktop 24 cores, albeit as E-cores, for a while, right?

And, old HEDT was a grunty 6 cores back in the day. Those certainly had the platform tax applied.

Also, I wouldn't class quad channel RAM as baggage either.
 
12 core, single chiplet, x3D CPU will be the perfect replacement for my 9800x3D by the time RAM prices are sane again in 2028.

8 cores are sufficient for most tasks, but there are times when I miss the extra cores of my 5900X.
by that point Zen 7 might be releasing, although maybe on a new socket
 
by that point Zen 7 might be releasing, although maybe on a new socket
Maybe, but I expect AI hording to slow everything down a bit. Also, there is usually a delay between announcement, actual release, and available at MSRP.

I don't mind a new socket as the 9800x3d + Mobo will move to my living room PC so I'll be getting a new mobo either way. But if the new socket also heralds the move to DDR6 and PCIe 6 that could make it less desirable from a value and stability standpoint. That's the same reason I waited for the 9000 series to move to AM5.
 
7 Ghz... Reminds me of that really bad period in Intel's past when they were ramping up the GHZ around the end of the Prescott days. Some really poor CPU's came out or got cancelled that were hot, incredibly inefficient and actually quite slow. Then they pivoted to Core 2 with all it's efficiency over GHZ and dominated the CPU space for a decade.
 
It means that the server side (Epyc) will likely start having up to 192 cores and 384 usable threads.

It's insane the push AMD has been doing.
That already exists? AMD Epyc 9965 is 192 Core, 384 Threads. It was released in October 2024, so it's already over a year old and getting closer to two years old.

The rumours for Epyc and Zen 6, is that it's going to hit 256 cores, 512 threads.
 
7 Ghz... Reminds me of that really bad period in Intel's past when they were ramping up the GHZ around the end of the Prescott days. Some really poor CPU's came out or got cancelled that were hot, incredibly inefficient and actually quite slow. Then they pivoted to Core 2 with all it's efficiency over GHZ and dominated the CPU space for a decade.

Exactly my thinking here too. It is mathematically the case that clockspeed increases offer diminishing returns. Going from 3Ghz to 4Ghz offers a bigger performance boost than 5Ghz to 6Ghz, while power consumption and heat tend to increase exponentially.
 
Back