Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[yuri69]

AMD does this with mobile parts quite often. Unsure why folks are making this a big deal, this has been happening for years.

There seems to be a nuance. Last time they relied a whole year solely on a rebrand was back in cash-strapped Zen 1 days with Raven Ridge/Picasso.

 

[SteinFG]

There seems to be a nuance. Last time they relied a whole year solely on a rebrand was back in cash-strapped Zen 1 days with Raven Ridge/Picasso.

7040 zen 4 phoenix to 8040 zen 4 hawk is also that - reuse of the same silicon.

 

[poke01]

I mean 2026 has nothing compelling from Intel anyway, Strix point refresh is fine.

Panther lake CPU seems to be a low key update.

 

[yuri69]

7040 zen 4 phoenix to 8040 zen 4 hawk is also that - reuse of the same silicon.

How is that the same situation?

* Phoenix launched in April 2023
* Hawk Point refresh launched in December 2023
* Strix Point launched in June 2024

Compare the Zen 5 situation:
* Strix Point launched in June 2024
* Gorgonzola refresh is scheduled to 2025-2026
* Medusa Point launches in Q1 2027

 

[Joe NYC]

Is Medusa Halo Zen5?

View attachment 120766

Medusa is Zen 6

 

[Joe NYC]

It's fairly typical for an AMD refresh, they just speed binned the highest end parts.

I wonder if AMD might be using the reduced cost N4C, at least the low end dies.

 

[Joe NYC]

How is that the same situation?

* Phoenix launched in April 2023
* Hawk Point refresh launched in December 2023
* Strix Point launched in June 2024

Compare the Zen 5 situation:
* Strix Point launched in June 2024
* Gorgonzola refresh is scheduled to 2025-2026
* Medusa Point launches in Q1 2027

Imagine there only 2 times of year: start and mid, and round to those:

Zen 4 desktop: Mid 2022
Zen 4 laptop Phoenix + 6 months - start 2023
Zen 4 laptop refresh Hawk +12 months - start 2024

Zen 5 desktop: + 24 months Mid 2024
Zen 5 laptop Strix + 0 months Mid 2024 (early)
Zen 5 laptop rest + 6 months - Start 2025
Zen 5 refresh + 12 months - Start 2026

Now, projecting to Zen 6
Zen 6 desktop: + 24 months, Mid 2026
Zen 6 laptop Medusa Point + 0 months Mid 2026
Zen 6 laptop rest+ 6 months - Start 2027

 

[fastandfurious6]

yep they're moving at breakneck speed

Yeah, but at what power limit? A lot of bigger 7945HX laptops allow 165W PPT, which for Zen 5 is only 35W below desktop 9950X(3D) at 200W PPT. Set a desktop 9950X(3D) to 165W PPT and I bet you only lose around 10% performance.

Not to mention, at least on Dragon Range, it ran JEDEC ram which meant the IOD power was very low, freeing up more power for the cores compared to desktop running EXPO. It also does funny stuff like clocking FCLK, UCLK, and MCLK down when under heavy CPU load that isn't doing heavy memory access, to free up even more power for the cores.

I think the ~10% avg will hold in other chassis too

 

[yuri69]

Zen 6 laptop Medusa Point + 0 months Mid 2026
Zen 6 laptop rest+ 6 months - Start 2027

AFAIK there is no leak supporting this idea.

It seems the Medusa Point would be Q1 2027 followed by the rest of mobile later.

 

[SteinFG]

AFAIK there is no leak supporting this idea.

It seems the Medusa Point would be Q1 2027 followed by the rest of mobile later.

it's a 2026-2027 prediction, obviously there's no leaks yet. I bet even AMD doesn't know. Summer release worked fine for strix point, It can happen again. Release High-end in limited quantities in 2026 Q3-Q4, increase production and introduce lower end models in 2027 CES

 

[Ghostsonplanets]

it's a 2026-2027 prediction, obviously there's no leaks yet. I bet even AMD doesn't know. Summer release worked fine for strix point, It can happen again. Release High-end in limited quantities in 2026 Q3-Q4, increase production and introduce lower end models in 2027 CES

Pretty sure Strix was mid-24 due to Copilot+.

 

[adroc_thurston]

ummer release worked fine for strix point, It can happen again

the cadence is fixed.

Pretty sure Strix was mid-24 due to Copilot+.

it was mid-24 because it was redefined like a zillion times.

 

[Joe NYC]

AFAIK there is no leak supporting this idea.

It seems the Medusa Point would be Q1 2027 followed by the rest of mobile later.

As far as leaks, while not giving explicit dates, MLID talks about Medusa as 2026, implying Mid 2026.

2nd is, as @adroc_thurston says, cadence. The original goal has been 18 months, but it has slipped to ~24 months.

So, what you are suggesting that it will slip further to 30 months. There is nothing that would point to Zen 6 (as a core) and Zen 6 as a client CCD that would be high risk to blowing up the schedule. Quite the opposite. Few things that could have been high risk that are, instead low risk to schedule slippage:
- Zen 6 core is only incremental to Zen 5
- ring bus upgrade (to mesh) to support more cores - done
- mobos, socket - reusing existing ones - done
- new connectivity between chiplets - done in Strix Halo
- new IODs - template done in Strix Halo

 

[yuri69]

As far as leaks, while not giving explicit dates, MLID talks about Medusa as 2026, implying Mid 2026.

2nd is, as @adroc_thurston says, cadence. The original goal has been 18 months, but it has slipped to ~24 months.

So, what you are suggesting that it will slip further to 30 months. There is nothing that would point to Zen 6 (as a core) and Zen 6 as a client CCD that would be high risk to blowing up the schedule. Quite the opposite. Few things that could have been high risk that are, instead low risk to schedule slippage:
- Zen 6 core is only incremental to Zen 5
- ring bus upgrade (to mesh) to support more cores - done
- mobos, socket - reusing existing ones - done
- new connectivity between chiplets - done in Strix Halo
- new IODs - template done in Strix Halo

A Zen 6-based product should come late 2026 but, apparently, not the mobile Medusa Point. There is Zen 6-based Olympic Ridge (and/or Venice?) to fill that 2026 release window.

As mobile is a notable source of income for AMD I've been a bit worried about that "rebrand-only" strategy.

 

[Joe NYC]

A Zen 6-based product should come late 2026 but, apparently, not the mobile Medusa Point. There is Zen 6-based Olympic Ridge (and/or Venice?) to fill that 2026 release window.

3 products should be chiplet based: Ridge, Point, Halo. Which imply the same CCD but 3 different IODs.

If the 24 month cadence is maintained, then 2 out of 3 need to be released in Mid 2026. Desktop (Ridge) and laptop (Point). Otherwise, it would be a slip. These two are also likely the most profitable products to have for 2026 Christmas season.

As mobile is a notable source of income for AMD I've been a bit worried about that "rebrand-only" strategy.

I speculated in another thread that AMD might be using cost reduced N4C process node, for the refresh, which, IMO, would be optimal strategy, with assumption that very strong chiplet based products are coming in 2026.

I think your worry about "rebrand only" would be justified, if Zen 6 is not released in Mid 2026. It would open the door wide open to Intel, to stop and reverse market share gains.

IMO, if Zen 6 really were to be a 2027 product, AMD would be pressed to be releasing N3 based refresh of Zen 4 laptops. Absence of such a product, IMO, is indication of Zen 6 taking the baton from Zen 5 right on time in 2026.

 

[adroc_thurston]

If the 24 month cadence is maintained, then 2 out of 3 need to be released in Mid 2026

They aren't.

 

[Ghostsonplanets]

Mobile cadence is tied to OEMs. Return to School period and CES are the biggest yearly seasons to target new Mobile SoCs.

 

[fastandfurious6]

I guess TSMC 2nm is the #1 staller

 

[yuri69]

I think your worry about "rebrand only" would be justified, if Zen 6 is not released in Mid 2026. It would open the door wide open to Intel, to stop and reverse market share gains.

IMO, if Zen 6 really were to be a 2027 product, AMD would be pressed to be releasing N3 based refresh of Zen 4 laptops. Absence of such a product, IMO, is indication of Zen 6 taking the baton from Zen 5 right on time in 2026.

I hoped for a N3 refresh filling the 2026 void but... AMD does irrational things like wasting their resources with "XT" desktops CPUs.

I guess TSMC 2nm is the #1 staller

AMD was so afraid of 3nm they split the lineup of Zen 5. Suddenly with Zen 6 they are willingly about to wait for acceptable 2nm yields with the whole lineup.

 

[fastandfurious6]

AMD was so afraid of 3nm they split the lineup of Zen 5. Suddenly with Zen 6 they are willingly about to wait for acceptable 2nm yields with the whole lineup.

it's a more complex decision making process than just this pattern comparison

zeitgeist, timings, competition, perf/w/thermals/die improvement ratios, estimated market reactions...

2nm is 100% worth the wait for medusa

 

[adroc_thurston]

AMD was so afraid of 3nm they split the lineup of Zen 5. Suddenly with Zen 6 they are willingly about to wait for acceptable 2nm yields with the whole lineup.

It's not a yield thing.
TSMC is the king of raw and parametric yield anyhow.

 

[HurleyBird]

it was mid-24 because it was redefined like a zillion times.

That sounds extremely dysfunctional. The obviously un-optimal (in terms of area and performance) dual CCX config is what I'd expect if AMD made the decision to sacrifice quality on the alter of cadence, and not something I'd expect given the opposite. Or do you mean the Zen5 core rather than the layout?

 

[adroc_thurston]

That sounds extremely dysfunctional

You can blame MS for having a truly schizophrenic set of AI PC targets.

 

[APU_Fusion]

AMD is hedging their bets with server/data center as main focus along with trying to capture ai market share. I imagine zen 6 focus is server first focus? Desktop is distant second thought.

 

[adroc_thurston]

I imagine zen 6 focus is server first focus?

Quite literally everything Zen since 2017 is "server focus".