Discussion Zen 7 speculation thread

[511]

IMEC Has A7 node being the first gen of CFET which is due about 2031 (probably 2032 realistically) and subsequently A5 in 2033 (again 2034 most likely).

Hold on a second here IMEC Roadmap is not practical for commercialization . It says N2 2024 and A14 2026 we are gretting N2 in 2025-6 and A14 2028-9

 

[dangerman1337]

Hold on a second here IMEC Roadmap is not practical for commercialization . It says N2 2024 and A14 2026 we are gretting N2 in 2025-6 and A14 2028-9View attachment 128779

They have a more recent one:

https://www.eetimes.com/vlsi-2025-outer-wall-forksheet-bridges-nanosheet-and-cfet-architectures/

https://www.imec-int.com/sites/default/files/2025-01/Figure%201%20-%20Imec%20roadmap.png

So yeah A7 being HVM hopefully in 2032 is plausible.

 

[Doug S]

So yeah A7 being HVM hopefully in 2032 is plausible.

Meh, I think we have to accept we're probably moving to three year cycles. With nodes getting more expensive more and more customers are going to want to amortize their design/mask costs for longer periods, and the gains of a "major node" continue to shrink. So instead of major node at time+0y, followed by a 'P' version at time+1y, followed by the next major node at time+2y, we'll start to see two 'P' type nodes before the next major node.

This is arguably already the case and we just haven't acknowledged it yet. Apple is doing its third iPhone with an N3 family node, and N2 ships in volume in 2026 and all indications are that A14 won't ship in volume until 2029.

IMEC doesn't exactly have a good track record with their roadmaps. Still waiting on 450mm wafers lol

 

[fastandfurious6]

by 2030 computing will have radically changed (or starting to change etc)

 

[igor_kavinski]

by 2030 computing will have radically changed (or starting to change etc)

How has it radically changed since 2020?

 

[fastandfurious6]

How has it radically changed since 2020?

two from the top of my head:

laptop = desktop

mobile phone upgrades = meaningless


basically chips got extrmly performant at 0-100w and everything extra is mostly score

0-50w chips run everything sufficiently, cool, fast, no lag

etc. big thermal/w optimization era

 

[lamedude]

Unreal Engine won't stutter in 2030

 

[adroc_thurston]

How has it radically changed since 2020?

we're finally off Skylake for one.

 

[igor_kavinski]

But there's nothing radical that's happened. No new paradigm shift. CPUs haven't learned any new tricks.

No SMT4, for example.

6 GHz was breached but not for mainstream CPUs. That should obviously change in 2026. Maybe 8 GHz by 2030 if we are very lucky?

APX looks great on paper and hoping it debuts in Zen 7 at least.

I think the next radical shift in computing will be when a single thread's multiple branches can be processed in parallel on different cores and instead of flushing the pipeline on the wrong branch prediction, the context simply changes to the core processing the correct branch.

 

[adroc_thurston]

But there's nothing radical that's happened. No new paradigm shift. CPUs haven't learned any new tricks.

Well, uh, yeah.
Normal ass cores won eons ago when speed daemons died an untimely death at the hands of Dennard Scaling.

No SMT4, for example

POWER has it and Vulcan and derivatives (Triton) had it too. But it's not worth much.

6 GHz was breached but not for mainstream CPUs. That should obviously change in 2026. Maybe 8 GHz by 2030 if we are very lucky

Means to an end.

APX looks great on paper

It doesn't.
It's aa64 thingies while missing the very point of why aa64 exists (cleaning up the ISA).

I think the next radical shift in computing will be when a single thread's multiple branches can be processed in parallel on different cores and instead of flushing the pipeline on the wrong branch prediction, the context simply changes to the core processing the correct branch

I just love extending the sidechannel attack surface even further beyond.
What could possibly go wrong?

 

[511]

we're finally off Skylake for one.

Skylake doesn't die

 

[igor_kavinski]

I just love extending the sidechannel attack surface even further beyond.
What could possibly go wrong?

Put the responsibility on the user.

If the user is not sure if some malicious process is running on the CPU, then don't use the feature.

Spectre/Meltdown was much ado about nothing. Is there a single widespread instance of operating systems running on unpatched CPUs getting compromised?