Question Zen 6 Speculation Thread

[CakeMonster]

What if launch in 2027 is because they are waiting for TSMC 2nm? I mean, Zen 5 was supposed to be a revolution some years ago and probably part of that was being on 3nm with the added transistor budget. Maybe AMD doesn't want to do the same mistake?

With the current cost of nodes, I think if Z6 is 2027 it is partly because its 3nm, not 2nm.

NV is doing the 'old' 4nm for consumer Blackwell in 2025 because of cost, so AMD will absolutely not pay for the bleeding edge.

 

[igor_kavinski]

Most likely Zen6 on N3 and Zen6c on N2.


What questionable features did bulldozer have?
View attachment 111343

Can you please share the original URL?

 

[Meteor Late]

With the current cost of nodes, I think if Z6 is 2027 it is partly because its 3nm, not 2nm.

NV is doing the 'old' 4nm for consumer Blackwell in 2025 because of cost, so AMD will absolutely not pay for the bleeding edge.

Hmm, I don't know, it seems weird to me then how Apple is upgrading to 16GB in base models and still maintaining prices while being at 3nm then. It cannot be THAT expensive.

 

[gdansk]

Can you please share the original URL?

transcript https://www.computerhistory.org/collections/catalog/102809019
video

 

[igor_kavinski]

how Apple is upgrading to 16GB in base models and still maintaining prices while being at 3nm then.

Answer is simple: it probably didn't cost them anything. Rest of the computer industry is moving towards denser chips. Less dense chips are getting phased out. It's possible that memory makers told Apple that they won't be able to supply them lower density chips in volume anymore. Apple is not going to give more out of the kindness of their heart (not that there is any kindness inside that block of stone anyway).

 

[yuri69]

video

omg, bring the man back to AMD.

 

[igor_kavinski]

omg, bring the man back to AMD.

He might return yet.

AMD may buy Tenstorrent

But AMD's terms may have a non-negotiable clause: "Raja Koduri is not welcome".

 

[OneEng2]

Going 2 ALUs, 16k L1d & 2-way L1i, horribly slow L2, and doing FMA in 6 cycles?

It failed to meet a competitive performance bar and as a result, missed every business objective it was supposed to meet driving the company nearly out of business.

On topic, I think we will see Zen 6 only if Intel produces something better than Zen 5. It's sad, but the way the world works 😔

 

[igor_kavinski]

AMD’s Zen 6 “Medusa” Ryzen CPUs: AM5 support confirmed, market launch planned for the end of 2026 | igor´sLAB

AMD’s upcoming Zen 6 generation Ryzen desktop CPUs are rumored to be compatible with the current AM5 socket and are expected to be launched in late 2026 or early 2027. According to initial reports…

www.igorslab.de

Has he lost his mind? Nowhere in the article is there any confirmation of Zen 6 being on AM5

 

[gdansk]

Hmm, I don't know, it seems weird to me then how Apple is upgrading to 16GB in base models and still maintaining prices while being at 3nm then. It cannot be THAT expensive.

Yep, cost isn't the issue. Apple is including ~160mm2 N3 silicon in a $600 Mac Mini. And it's Apple so you know they have better margins than AMD. It's also an entire computer with an internal power supply. A Ryzen 9 AI 495X with 2 x N3 CCDs is probably only 140mm2...

 

[FlameTail]

If Zen6 is using 3nm, which 3nm node will it specifically use? There are many.

N3B
N3E
N3P
N3X

A late 2026 release suggests that it could use any of these, because all four of them would be in mass production by that time.

It's most likely N3P or N3X though.

N3X is a substantial upgrade over N3P. There's even a 1.1x density improvement*** over N3P, which in turn is a 1.04x density improvement over N3E.

	Power	Density
N3E	100	100
N3P	95/90	104
N3X	88.5/83.7***	114.4***
N2	75/70	115

N3X is surprisingly close to N2.

***Only at the same performance level.

 

[StefanR5R]

Has he lost his mind?

FYI, the Samir Bashir articles at igorslab.de are just filler, badly informed, and can be safely ignored.

 

[OneEng2]

If Zen6 is using 3nm, which 3nm node will it specifically use? There are many.

N3B
N3E
N3P
N3X

A late 2026 release suggests that it could use any of these, because all four of them would be in mass production by that time.

It's most likely N3P or N3X though.
View attachment 111356
N3X is a substantial upgrade over N3P. There's even a 1.1x density improvement*** over N3P, which in turn is a 1.04x density improvement over N3E.

	Power	Density
N3E	100	100
N3P	95/90	104
N3X	88.5/83.7***	114.4***
N2	75/70	115

N3X is surprisingly close to N2.

***Only at the same performance level.

I am betting that it is not N3 of any flavor but rather N2.

It isn't the density that I think is the big problem, it is the power. Remember, Zen 6 will be getting 32 core CCD's. They are going to need the best thermals possible to fit as many CCD's as possible within the DC socket power limit. The DC is the key to AMD having plenty of cash to build on their success. I know we all fixate on Desktop (and even gaming) here, but from a strategic standpoint, it really is all about design for power efficiency. This is true for DC due to the socket power limits, and it is true for laptops because of battery life.

I think that N2 will still be a pricey node for AMD in H2 2026, but it will be a necessary node for DC IMO. Perhaps they will stick to N3P/N3X for desktop/laptop CCD's. It seems like AMD has had pretty good success with Zen 5 on 2 different nodes, so they seem to have the design tools to pull this off fairly well.

 

[Thibsie]

Zen 6 will have 32 cores CCD for server parts.
AFAIR desktop will be based on mobile, not server anymore.

 

[Doug S]

Hmm, I don't know, it seems weird to me then how Apple is upgrading to 16GB in base models and still maintaining prices while being at 3nm then. It cannot be THAT expensive.

Sure AMD can easily handle the cost of N3E on their higher end stuff, but that's not where the bulk of their sales are. Apple's cheapest N3B/N3E laptop is $1099. What percentage of AMD laptops would you say sell for more than that? I'd guess maybe 10-15%. They're also competing against Intel, who isn't exactly hitting on all cylinders lately so why should AMD increase their cost when it wouldn't benefit them competitively? They always have the option to do a version of Zen 5 on N3E later, but why rush as long as they are selling plenty of them on a cheaper process and making more money?

The situation with Apple is totally different. The only Mac you can buy for under $1000 is the Mini, which is kind of an outlier price wise. It may be sort of the "iPhone SE" of the Mac line, something that's cheap enough some people who are in the Windows world are willing to say "eh, I'll buy one to play around with" and if Apple converts 5 or 10% of those "Mac-curious" customers it is well worth selling one model at lower margins than the rest of the Mac line.

 

[Thibsie]

[...]and if Apple converts 5 or 10% of those "Mac-curious" customers it is well worth selling one model at lower margins than the rest of the Mac line.

Yeah their margin on Mac Mini M4 RAM and SSD or so bad they might get poor quickly

 

[Doug S]

Yeah their margin on Mac Mini M4 RAM and SSD or so bad they might poor quickly

Its not that it is bad, just that it can't be as high as it is on the Mac line overall. Same with the iPhone SE.

 

[Thunder 57]

It's a shrink of Zen 1 with improved clock rates that took 13 months to follow Zen.
AMD already launched the N3E shrink of Zen 5C only months after the N4P version and you're saying they're less nimble?
The problem is cadence. Zen 5 is a Q1 2024 product that launched in Q3. Zen 6 is a competitive product in Q3 2025 but it launches in... 2027?

Nobody with a brain would think 2027. I don't care what some shmuck might say. That would be buffoonery.

 

[poke01]

Okay who cares about cadence?

We got Apple, Intel and ARM who release chips every year because they make SMALL improvements to their core

Apple - M1 to M3 few updates, clock increases. Different ecosystem
Intel - don't even need to explain their stupidity
ARM - Don't be fooled by IPC improvements their clocks are still too low to feasible on desktop/laptop
Qualcomm - This is the only threat if their V3 core is good but even then it runs on arm64. So its not for desktop gamers or companies who need use on x86 software. laptop is the only market they can take which AMD doesn't services.

AMD is doing fine even on 16-22 month cadence. Its not how much you output but the quality of the releases. Zen5 is massive for AMD and Zen6 will be no different. FixFlex will help a lot on desktop and so will the new IOD.

 

[gdansk]

ARM - Don't be fooled by IPC improvements their clocks are still too low to feasible on desktop/laptop

X925 is ~15% lower 1T performance than Intel and AMD's newest desktop chips. In a phone. It has similar performance to their laptop chips.
X930 will match GR/ARL desktop. In a phone! and exceed their laptop chips.
They're going to lose one of the last two volume markets that x86 has - laptops.
Microsoft is prioritizing WoA for a reason.

We already know Zen 6 is only about ~10% IPC. It simply won't be competitive in the consumer space. Even Intel can beat that unless they implode.

 

[poke01]

X925 is ~15% lower 1T performance than Intel and AMD's newest desktop chips. In a phone. It has similar performance to their laptop chips.
X930 will match GR/ARL desktop. In a phone! and exceed their laptop chips.
They're going to lose one of the last two volume markets that x86 has - laptops.
Microsoft is prioritizing WoA for a reason.

I think they won't. There is more to the x86 ecosystem than IPC.

 

[gdansk]

I think they won't. There is more to the x86 ecosystem than IPC.

Yeah, like bad battery life and less competition. Qualcomm and ARM are designing these cores every year anyway for the phone market. It adds to their TAM by pushing into laptops now that they can match the performance of Strix Point/Lunar Lake/Arrow Lake and exceed it next year.

 

[adroc_thurston]

X925 is ~15% lower 1T performance than Intel and AMD's newest desktop chips. In a phone. It has similar performance to their laptop chips.

I put Strix on a 3-2 N3e and now it's 30% faster.
Not iso node, not relevant/fair/yaddayadda.

 

[gdansk]

I put Strix on a 3-2 N3e and now it's 30% faster.
Not iso node, not relevant/fair/yaddayadda.

And when is AMD doing that?

 

[adroc_thurston]

And when is AMD doing that?

Medusa mobile! But it'll be N3e and in early 2027.