Question Zen 6 Speculation Thread

[adroc_thurston]

ofc should i announce that N5/4 is worse than Intel 7 cause apparently they can't match the clocks of a wider core ?

Safe RPLs clock 5.6 and lower which is slower than N5p.

 

[Markfw]

Safe RPLs clock 5.6 and lower which is slower than N5p.

Can we stop talking about nova lake in a Zen 6 thread ?

 

[OneEng2]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

If NVL can make a convincing case against a refreshed Zen5 then I would propose that AMD jumped the shark with betting on such an advanced node for desktop.

 

[MoistOintment]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

If NVL can make a convincing case against a refreshed Zen5 then I would propose that AMD jumped the shark with betting on such an advanced node for desktop.

Makes no sense to me. AMD is willing to essentially "delay" Zen 6 a year for N2X instead of just using N2 or N2P? How does that financially make sense? And then Zen 6 is a Razer Lake (or whatever is after NVL) competitor? I don't buy it.

 

[poke01]

so the desktop market amd don’t give a crap about is going to use N2X as per MILD…

 

[marees]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

If NVL can make a convincing case against a refreshed Zen5 then I would propose that AMD jumped the shark with betting on such an advanced node for desktop.

Is this what you are saying ?

Zen 6 desktop chips may arrive in late 2026, following APU and EPYC rollouts - Moore's Law Is Dead

AMD's Zen 6 desktop CPUs, based on the Olympic Ridge platform and N2X process, may not launch until late 2026. This would place them after the release of N2P-based EPYC server processors and Medusa Point APUs, both expected in the first half of that year.

www.notebookcheck.net

Zen 6 desktop chips may arrive in late 2026, following APU and EPYC rollouts - Moore's Law Is Dead​

AMD's Zen 6 desktop CPUs, based on the Olympic Ridge platform and N2X process, may not launch until late 2026. This would place them after the release of N2P-based EPYC server processors and Medusa Point APUs, both expected in the first half of that year.

Olympic Ridge, which uses the cutting-edge N2X node, is positioned as the final phase of this rollout. While some mainstream desktop CPUs based on N2P could appear earlier—possibly by August 2026—the flagship N2X models may follow closer to the end of the year. AMD’s choice to lead with APUs and EPYC chips suggests a prioritization of volume and mobile/server platforms ahead of ultra-high-performance desktop parts. ​

 

[Thunder 57]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

If NVL can make a convincing case against a refreshed Zen5 then I would propose that AMD jumped the shark with betting on such an advanced node for desktop.

Zen 6 in 2027 would be an epyc fail. I'm not buying it.

 

[marees]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

If NVL can make a convincing case against a refreshed Zen5 then I would propose that AMD jumped the shark with betting on such an advanced node for desktop.

Worth separating his leaks from his spin/fiction

Nodes I believe are not set in stone & AMD / TSMC can change — if not already taped out

Refresh (of zen 5) is extremely unlikely — most likely co-hallucinated with chatGPT

 

[vanplayer]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

Zen6 is never planned to be on N2X. It is only on N2P or worse. Also MLID's NVL projection is also way off.

 

[reaperrr3]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

MLID sometimes spins wrong conclusions out of correct info.

Keep in mind modern processes have both different transistor variants and other parameters that can be tweaked per production line.
AMD's FX 9590 (Centurion) was just Vishera produced with some customized parameters geared towards higher clocks at the price of more leakage, but the process as such was otherwise still the same 32nm.

Big customers like AMD get some preferrential treatment at TSMC, including involvement during process R&D and tailoring some parameters on their booked production lines to suit their needs.

The 12C CCD will certainly be produced on multiple production lines at once, wouldn't surprise me if some of those were configured a bit more towards power efficiency (for mobile and maybe server), while some other lines have their parameters tweaked a bit more towards allowing a bit more leakage, voltage and therefore higher maximum clocks (for deskop).
So it might not be "true" N2X, but desktop CCDs may get manufactured with some "X"-like parameters to get nearly the same effect.

So no, I don't expect the N2X roadmap delays to affect Zen6 much, if at all.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

Yeah.
N2P only promises like 3% more perf at iso power over vanilla N2 anyway, lowest "P" improvement ever so far, so definitely not worth risking a delay of a server product over, where potential customers probably care a lot about timely releases they can plan for.

In fact, this could be the other reason for the Verano refresh in 2027:
Not just a new IOD, but also a small process bump to N2P for the CCDs.

 

[burninatortech4]

Isn't the whole point of Zen die reuse? Why would Zen 6 compute dies for desktop and EPYC be on different processes?

Release timeline usually has more to do with prioritizing EPYC volume than anything else.

 

[DrMrLordX]

So ..... Really? Does everyone else here buy into this?

No.

 

[reaperrr3]

Isn't the whole point of Zen die reuse? Why would Zen 6 compute dies for desktop and EPYC be on different processes?

Release timeline usually has more to do with prioritizing EPYC volume than anything else.

We're talking about the 32c dense server die being on N2 and the 12C die being on N2P.

As for the parameter thing, that's not really the same as being on a different process, but it's just my speculation anyway.

 

[Josh128]

I guess I must have been sleeping.

MLID is saying:

1) Zen 6 is on N2X
2) N2X is delayed until 2027 (TSMC slide)
3) Therefore Zen 6 is delayed until 2027
4) AMD will use X3D2 and a refresh of Zen 5 to hold off NVL

So ..... Really? Does everyone else here buy into this?

I guess if I were AMD ..... AND I knew NVL would not supplant a refreshed Zen 5 in performance (excluding MT), perhaps I would keep on N4P and rake in profits waiting on Zen 6 .... but still, that is an insane delay if true.

I'm still betting Venice is released in 2026 (H1) on N2 (not N2P).

If NVL can make a convincing case against a refreshed Zen5 then I would propose that AMD jumped the shark with betting on such an advanced node for desktop.

Indeed AMD has publicly said Venice is coming in 26, everything else we've seen thus far, including mobile and DT, points to 2027. I recall arguing with Adhoc about this, telling him Venice is coming first. That said, I dont expect to see it before July or August at the earliest. If it actually makes H1, its going to be right at the deadline.

Zen 5 matched Arrow Lake 4nm vs 3nm. Thats not going to happen if it has to combat NVL on 2nm, regardless of a refresh. NVL looks like a big update from Intel complete with AVX10 and other HPC goodies. If Intel cant manage to beat a 2 year old 4nm design on 2nm, they are truly cooked.

That said, in gaming specifically, I dont expect NVL would beat a hypothetical 9850X3D by any appreciable amount, perhaps not even 9800X3D, and its certainly not going to be worth it for gaming for the price it will require.

 

[Josh128]

No.

Yes. I think its quite likely if the whole "muh GHz" thing is true. You aint getting world record frequencies on a standard iphone/server class node.

 

[OneEng2]

We're talking about the 32c dense server die being on N2 and the 12C die being on N2P.

As for the parameter thing, that's not really the same as being on a different process, but it's just my speculation anyway.

I can understand waiting on N2P only if AMD believes they need it to combat NVL for the next generation (ie, vanilla N2 wouldn't cut it). Just N2P puts Zen 6 desktop off to the end of 2026 to the beginning of 2027. N2X? Now you are off into the ether.

On the other hand, AMD did NOT jump on the first gen N3B node. They waited until N3E and even then only used it for high margin DC chips.

I could still see an AMD Zen 6 desktop/mobile part on N3P. Such a part would likely be more than enough to combat NVL (just as Zen 5 on N4P is more than enough to combat ARL on N3E). Such a chip could be released at AMD's leisure without waiting for a new node to mature.

 

[adroc_thurston]

I can understand waiting on N2P only if AMD believes they need it to combat NVL for the next generation (ie, vanilla N2 wouldn't cut it). Just N2P puts Zen 6 desktop off to the end of 2026 to the beginning of 2027. N2X? Now you are off into the ether.

They don't 'wait' for anything since neither N2p nor N2x exist for AMD.

On the other hand, AMD did NOT jump on the first gen N3B node.

Indeed. It exploded.

I could still see an AMD Zen 6 desktop/mobile part on N3P. Such a part would likely be more than enough to combat NVL (just as Zen 5 on N4P is more than enough to combat ARL on N3E).

Can you stop the dementia patient bit? Ze FAD chart says N2 and that's exactly what you're getting.
Oh and AMD cares not about Intel. Not one bit.