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

Page 298 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

soresu

Platinum Member
Dec 19, 2014
2,662
1,862
136
I thought Zen5 was on TSMC 4nm?
Given the history of Zen cores so far I wouldn't expect them to be tied down to any single node.

Zen1 had 2 nodes (at least), Zen2 had 2 (possibly more depending on what node PS5 Pro SoC is fabbed on), Zen3 had 2 and we're probably still not at the end of Zen4's lifecycle yet, so it may be on more than 1 node also.

I wouldn't at all be surprised to find Zen5 being fabbed on more than one node during its lifecycle considering its significant departure from current core design.
 

biostud

Lifer
Feb 27, 2003
18,250
4,763
136
Given the history of Zen cores so far I wouldn't expect them to be tied down to any single node.

Zen1 had 2 nodes (at least), Zen2 had 2 (possibly more depending on what node PS5 Pro SoC is fabbed on), Zen3 had 2 and we're probably still not at the end of Zen4's lifecycle yet, so it may be on more than 1 node also.

I wouldn't at all be surprised to find Zen5 being fabbed on more than one node during its lifecycle considering its significant departure from current core design.
I was merely considering the launch desktop node, not laptop, dense cores or future revisions of the design. :)
 
Jul 27, 2020
16,307
10,333
106
Straight from the horse's mouth (you may call him AMD):

5950X: TSMC 7nm FinFET
7950X: TSMC 5nm FinFET

5900HS: TSMC 7nm FinFET
6900HS: TSMC 6nm FinFET
7940HS: TSMC 4nm FinFET
8945HS: TSMC 4nm FinFET

5900HX: TSMC 7nm FinFET
6900HX: TSMC 6nm FinFET
7945HX: TSMC 5nm FinFET


5850U: TSMC 7nm FinFET
6850U: TSMC 6nm FinFET
7840U: TSMC 4nm FinFET
8840U: TSMC 4nm FinFET

7545U (with Zen4c cores): TSMC 4nm FinFET
 

soresu

Platinum Member
Dec 19, 2014
2,662
1,862
136
Straight from the horse's mouth (you may call him AMD):

5950X: TSMC 7nm FinFET
7950X: TSMC 5nm FinFET

5900HS: TSMC 7nm FinFET
6900HS: TSMC 6nm FinFET
7940HS: TSMC 4nm FinFET
8945HS: TSMC 4nm FinFET

5900HX: TSMC 7nm FinFET
6900HX: TSMC 6nm FinFET
7945HX: TSMC 5nm FinFET


5850U: TSMC 7nm FinFET
6850U: TSMC 6nm FinFET
7840U: TSMC 4nm FinFET
8840U: TSMC 4nm FinFET

7545U (with Zen4c cores): TSMC 4nm FinFET
Ah so definitely N5 and N4 then.
 

Timmah!

Golden Member
Jul 24, 2010
1,419
631
136
Yep. This use of the prefix "iso" to mean "same" is very common in engineering. I typically deal with isothermal (same temperature), isotropic (same physical properties), isocontour (same line or surface on a plot), etc.
But bunch of us are not engineers (well sort of in my case) and we stopped dealing with the iso term the moment burning DVDs went out of fashion :p
 

PJVol

Senior member
May 25, 2020
534
447
106
AMD zen 3 ISSCC
Yeah, something is changed in an official/marketing language since then. This quote is from ISSCC 2020 / February 17, 2020:
A principal objective for Zen 2 was to deliver better performance/W compared to
the previous-generation design. Zen 2 silicon is running at frequencies
comparable to Zen (Fig. 2.1.7), while delivering substantial performance and
power improvements through a combination of architectural IPC enhancement,
physical design methodology, and macro improvements which exploit 7nm
technology.

P.S. As a side note, purely tech docs
from PPR's for 17h (2020)
1709122980145.png

and PPR 19h model 11h (2023) :)
1709123059537.png
 
Last edited:

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Consoles really don't need that, they're not laptops.
It's for always online background downloads stuff without the usual horrible hacks.

It's a tricky challenge yeah.

Yea it's the same and very basic still.
Fancypants packaging is all Venice gen parts.
So current Genoa Epyc IO die has 16x GMI with 4 not used in any current parts?
 

cortexa99

Senior member
Jul 2, 2018
319
505
136

A korean site leak tons of insider materials and AMD's Zen5 APU lineup are also among these:

Source: https://gamma0burst.tistory.com/m/1107

download (1).png
AMD APU codename?

Strix A0(7nm) / Sarlak (7nm) / Strix B0 (7nm) / Kraken(5nm) / Sound wave(3nm)


Strix and Kraken are APU products confirmed in existing rumors, so other products are also assumed to be APUs.

Stirx A0, B0 are believed to be the rumored Strix Halo (chiplet) and Strix Point (monolithic).

I don't know which one is A0 or B0. (But there are also rumors that Sarlak is a Strix Halo.)

In the rumor, both Strix and Kraken are 4nm, so the difference from 7nm is large, so the reliability of the process indicated here should be considered low.

If you focus on the specifications, it is likely that they are all Zen5 and RDNA3.5 specifications.


download.png

Strix, Sarlak, Kraken IO dies.

It can be confirmed that the three products have a chiplet structure.


The product would have been marked as having a chiplet structure, but considering that Strix and Sarlak were marked separately, there is a high possibility that the rumor that Sarlak = Strix Halo is not true...

There is room for interpretation that the three products use the same IO die, but since the IO die also includes a GPU, looking at the GPU specifications alone, this seems unlikely.

Looking at the rumored roadmap, it is tied to the same lineup as monolithic Strix and Kraken, and the monolithic Strix GPU is 16CU.

The chiplet Strix GPU specifications are known to be 40CU, but if you look at the Kraken GPU specifications as 16CU, the difference is too large to cover with a cut chip.


How’d you know it’s coming out in April?
I guess he's just satirized the rumor mill about the Zen5 recently.:p


Anyway, RUMOR time:

1.the latest Zen5 revision/stepping is B2 (not very confident)

2. https://tieba.baidu.com/p/8916074174
ZEN5 desktop start mass packaging next month
dunno what is "packaging". OTOH, not only single source were complaining the Zen5's slower-than-expected progress, looks like there must be something happened. Some supply chain/factory did received orders related to Zen5 not too long ago but progress was slow. Don't ask the exact release date or perf number, I saw some "people in the know" also feel shocked why perf number is not leaked yet.
 

soresu

Platinum Member
Dec 19, 2014
2,662
1,862
136
Sound Wave seems to be a codename for a Zen6/RDNA5 APU according to that 'source'.

Something also to bear in mind (much as I despise mentioning it) that one of the various content spewing machines implied a rumour that RDNA5 µArch would be the foundation for future CDNA µArchs going forward, finally replacing the Vega/GFX9 foundation they have been using since CDNA1.
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,350
1,535
136
I saw some "people in the know" also feel shocked why perf number is not leaked yet.

If the rumor about it using the exact same IO die is correct, it's possible AMD feels that much less Zen5 specific development and testing is needed by the MB vendors. In principle, could be as far as AMD developing new AGESA, dropping it for MB vendors without any samples, and then testing their boards with the new chips themselves. There would just be dramatically fewer things that could go wrong than usual.

This would lead to less leaks; CPU leaks usually come from neither AMD nor Intel, but from their board partners (of which there are many, and who have much less incentive to run a tight ship).
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,777
247
106
That thread also says (after google-translating to English):

”It is predicted that ipc will increase by 19%, the same as zen3”

”Computex Taipei should be about zen5, that’s all it will be in the first half of this year, otherwise it will be another vest conference like CES.”

”The zen5 conference may be the release of Ryzen and Xiaolong at the same time. The Xiaolong will be distributed in advance, and the Ryzen will be delayed by 1-2 months.”


But I don’t know how much it can be trusted. Maybe they are just speculating, like is being done in this thread here.
 
Last edited:

cortexa99

Senior member
Jul 2, 2018
319
505
136
This is roughly speaking:
  • Bonding various dies to the substrate.
  • Adding the IHS on top.
  • Adding the traces for the LGA bumps.
  • Bonding the LGA bumps themselves.
Packaging is everything that isn't the actual semiconductor lithography in chip manufacturing.
You're right. Author's narrative feeling like it's the last step of a CPU being made, possibly IHS/LGA related.


That thread also says (after google-translating to English):
The most useful information were from thread's author, also the admin of that forum. The rest of the posts by others are just speculation.

Also something unrelated to Zen5, he seems mentioned R4(RDNA4?) is about to enter mass production mode very soon.
 
Last edited: