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

[DisEnchantment]

Speculate at will

 

[Joe NYC]

Not quite. They say that according to TechRadar AMD says that Zen 6 is on target for 2025.

From the linked TechRadar article;


A release could be anything... serverside included. And the way TechRadar mentions the 2025 date it sounds like it wasn't even part of AMD's Computex talking points, but referring to perhaps an earlier rumour.

"still on track for a release in 2025" - I'm scratching my head on this one... I can't remember any rumours to such an effect. MLID perhaps? An extrapolation of the Nirvana/Morpheus roadmap? Nirvana was 2023 - Morpheus 2024... so add a year to both... and hey you get a Zen 6 release in 2025. Centipede in action?

View attachment 103625

Just a lot of speculation all around. I added my 2 cents earlier when I speculated that long validation period for Zen 5 could mean shorter time interval to Zen 6.

And I came across these 2 articles, both referring to possibility of Zen 6 in 2025, but neither has a direct quote from AMD about expected timing.

There will be some events that may present an opportunity for AMD to narrow down the Zen 6 release date, such as a possible slide deck for upcoming Zen 5 reviews and upcoming Q2 earnings conference next week. But most likely, neither one will release anything new.

 

[Hail The Brain Slug]

I agree I think Zen 6 will tie into the release of DDR6.

However it doesn't mean we may not see some kinda of castrated version of Zen 6 for AM5 but doubting the 2025 release date.

There was that article semi-recently that I'm pretty sure said desktop DDR6 wouldn't be ready until 2027, so I think Zen 6 will be the last DDR5 generation.

 

[LightningZ71]

Has anyone thought that there might be a Zen 6 on an MI product, like mi450x or something like that? It would support the needed margins.

Maybe they will pull a Zen5+, kind of like Zen1+ on the 2x00 series? Maybe (puts on tinfoil hat) it could be the rumored Zen5 that was meant for N3 finally getting released?

 

[Joe NYC]

Has anyone thought that there might be a Zen 6 on an MI product, like mi450x or something like that? It would support the needed margins.

Maybe they will pull a Zen5+, kind of like Zen1+ on the 2x00 series? Maybe (puts on tinfoil hat) it could be the rumored Zen5 that was meant for N3 finally getting released?

It seems that AMD changed the names. Mi325 will likely be only HBM 3e, then Mi350, if there is an Mi350a version might have Zen 5 CPUs.

 

[adroc_thurston]

it could be the rumored Zen5 that was meant for N3 finally getting released?

Yeah it's called Turin Dense and it's on N3e.

desktop DDR6

What's that

 

[poke01]

there won’t be desktop DDR6. I think we’ll on to CAMM2

 

[Joe NYC]

What's that

Only LPDDR6 CAMM for desktop?

https://twitter.com/x/status/1815398474172535271

 

[poke01]

Only LPDDR6 CAMM for desktop?

https://twitter.com/x/status/1815398474172535271

Yep, that’s the future. It’s better than DIMMs

 

[branch_suggestion]

I agree I think Zen 6 will tie into the release of DDR6.

However it doesn't mean we may not see some kinda of castrated version of Zen 6 for AM5 but doubting the 2025 release date.

There was that article semi-recently that I'm pretty sure said desktop DDR6 wouldn't be ready until 2027, so I think Zen 6 will be the last DDR5 generation.

Desktop DDR6 if it is a thing will not be like the server version, more accurately I believe it will be LPDDR6 on CAMM.
Medusa being a unified client platform seems to be alluding convergence between laptop and desktop.

 

[adroc_thurston]

Only LPDDR6 CAMM for desktop?

Probably.
Mainstay DDRx has been steadily moving into server-memory-direction.

 

[RnR_au]

Yep, that’s the future. It’s better than DIMMs

There is also CUDIMM's coming. Suspect CAMM2 (or rather CAMM3 since JEDEC wants a new toolless version) will only be for laptops.

DDR5 vs DDR4: Is It Time To Upgrade Your RAM?

Welcome to the new era of DDR5 RAM

www.anandtech.com

 

[branch_suggestion]

There is also CUDIMM's coming. Suspect CAMM2 (or rather CAMM3 since JEDEC wants a new toolless version) will only be for laptops.

DDR5 vs DDR4: Is It Time To Upgrade Your RAM?

Welcome to the new era of DDR5 RAM

www.anandtech.com

That will extend existing DDR5 infrastructure for a while longer, but CAMM is just plain superior.
DDR6 has a big baseline cost associated and really is tailored for servers and nothing else.

 

[Goop_reformed]

Not hearing good news about zen 5 gaming performance atm sadly

 

[BorisTheBlade82]

In the linked article, Mike Clark indeed confirms, that they have 4 designs for Zen5 altogether, because they cut down AVX512 on mobile Zen5 and Zen5c:

For what we’re launching today in Strix Point, both the performance core and the compact core both have the AVX cut-down [AVX-256] because they're in a heterogeneous situation, and they're in a mobile platform where area is at a premium.

An interview with AMD's Mike Clark, the Father of Zen — 'Zen Daddy' says 3nm Zen 5 is coming fast; also talks compact cores for desktop chips

AMD expands on its Zen 5 architecture.

www.tomshardware.com

 

[SteinFG]

I just realised - Desktop Zen 3 and Zen 5 didn't require any unique silicon just for it. IO die for Zen 3 is reused from Zen 2, and the CCD is reused from / shared with Epyc.

Same with Zen 5 borrowing io die from Desktop Zen 4 and CCD from Epyc.

I know it's obvious but I didn't realise it until now hahah

Intel also goes in this direction. If Arrow Lake SOC tile turns out fine, they can use the same one for gen after arrow lake, and borrow 6P+8E tile from mobile. (though they still have to make 8+16 tile just for desktop)

 

[stayfrosty]

In the linked article, Mike Clark indeed confirms, that they have 4 designs for Zen5 altogether, because they cut down AVX512 on mobile Zen5 and Zen5c:

An interview with AMD's Mike Clark, the Father of Zen — 'Zen Daddy' says 3nm Zen 5 is coming fast; also talks compact cores for desktop chips

AMD expands on its Zen 5 architecture.

www.tomshardware.com

TH: What was the biggest challenge you encountered with Zen 5 development?

MC: It was actually dealing with two technologies [designing Zen 5 for both the 4nm and 3nm process technologies], especially a technology that the previous generation was in. And trying to do so much change, and therefore the unavoidable reality that in 4nm it's going to be [consume] more power than it's going to be in 3nm, no matter how smart we are.

But we need that flexibility in our roadmap, and it makes sense. But still that was really hard to try to control having the two technologies and the features, and a feature that looks great in 3nm not looking so great in 4nm because of the power impact of the not-as-efficient transistor and how it affects the floorplan. Normally, we do the architecture in one, and then we port on the next one, and then you have a lot of time to deal in the floor plan with the two technologies. [..] It was just really challenging. But that gives Zen 6 a lot of room to improve.
[..]

I wonder what they had to cut because of 4nm.

 

[fastandfurious6]

Finally the GB6 run with Strix boosting properly and it breaks the 2.9k score for the first time.

ASUSTeK COMPUTER INC. ProArt P16 H7606WI_H7606WI - Geekbench

Benchmark results for an ASUSTeK COMPUTER INC. ProArt P16 H7606WI_H7606WI with an AMD Ryzen AI 9 HX 370 processor.

browser.geekbench.com

ASUSTeK COMPUTER INC. ProArt P16 H7606WI_H7606WI - Geekbench

Benchmark results for an ASUSTeK COMPUTER INC. ProArt P16 H7606WI_H7606WI with an AMD Ryzen AI 9 HX 370 processor.

browser.geekbench.com

Spoiler: Frequency

"frequencies": [
5137,
5137,
5135,
5136,
4914,
4939,
5137,
5135,
5137,
5135,
5137,
5138,
5135,
5138,
5137,
5137,
5136,
5135,
5137,
5134,
5137,
5138,
5135,
5132,
5135,
5137,
5138,
5136,
5134,
5138,
5137,
5138,
5137,
5136,
5099,
5138,
5137,
5136,
5136,
5137,
5138,
5137,
5138,
4966,
5138,
5138,
5136,
5137,
5137,
5136,
5136,
5137
]

finally this is impressive, almost better than 7945hx with 4 less cores and presumably less watt and thermals

 

[fastandfurious6]

I wonder what they had to cut because of 4nm.

the mythical +36% IPC !!!

 

[MS_AT]

Not hearing good news about zen 5 gaming performance atm sadly

Anything more concrete? Is it below Zen4, below Zen1? Do you mean desktop or laptop flavor?

Anyway I was wondering are decoders in Zen5 statically partitioned between SMT threads, so when SMT threads is enabled each thread ends up with 4 wide decoder and only if you disable SMT in the BIOS it might happen that both decoders will be trying to decode the same instruction stream?

 

[burninatortech4]

I just realised - Desktop Zen 3 and Zen 5 didn't require any unique silicon just for it. IO die for Zen 3 is reused from Zen 2, and the CCD is reused from / shared with Epyc.

Same with Zen 5 borrowing io die from Desktop Zen 4 and CCD from Epyc.

I know it's obvious but I didn't realise it until now hahah

Intel also goes in this direction. If Arrow Lake SOC tile turns out fine, they can use the same one for gen after arrow lake, and borrow 6P+8E tile from mobile. (though they still have to make 8+16 tile just for desktop)

Not just Zen 3 and 5. That's been a core part of the Zen strategy from the beginning - reusing dies. All desktop and server dies(not laptop and excluding I/O dies) share the same silicon. Laptop chips are monolithic and, while sharing the same core architecture, have a different design philosophy (mobile).

 

[inf64]

Not hearing good news about zen 5 gaming performance atm sadly

CapFrameX as a source or something else? There is zero reasons for Zen 5 to have "bad" gaming performance.

 

[Philste]

Not hearing good news about zen 5 gaming performance atm sadly

CapFrameX as a source or something else? There is zero reasons for Zen 5 to have "bad" gaming performance.

AMD themselves basically confirmed that ZEN5 is bad at gaming. The Intel Comparisons were probably cherrypicked or some weird Profile for Intel ones. The "9700X is 12% faster than 5800X3D with 6000 EXPO RAM" comparison means that 9700X is 4-5% faster than 7700X. So ZEN5% is actually real in gaming.

Oh and btw Computerbase Editor just wrote this morning that there is no performance regression at all for 14900K and co when using the new profiles. Computerbase does a completely new parcours for both gaming and applications starting with ZEN5 so they habe to retest everything. He wrote, there is like a few specific situations where you can see a slightly lower score, but only if you know it. It's basically in the run to run variance region. So there's no 10% slower or something.

 

[fastandfurious6]

MLID has lost the plot, officially.

he gets paid for this!

 

[fastandfurious6]

AMD themselves basically confirmed that ZEN5 is bad at gaming. The Intel Comparisons were probably cherrypicked or some weird Profile for Intel ones. The "9700X is 12% faster than 5800X3D with 6000 EXPO RAM" comparison means that 9700X is 4-5% faster than 7700X. So ZEN5% is actually real in gaming.

Oh and btw Computerbase Editor just wrote this morning that there is no performance regression at all for 14900K and co when using the new profiles. Computerbase does a completely new parcours for both gaming and applications starting with ZEN5 so they habe to retest everything. He wrote, there is like a few specific situations where you can see a slightly lower score, but only if you know it. It's basically in the run to run variance region. So there's no 10 slower or something.

maybe new X3D improvements give the needed upper hand in gaming

 

[Nothingness]

Oh and btw Computerbase Editor just wrote this morning that there is no performance regression at all for 14900K and co when using the new profiles. Computerbase does a completely new parcours for both gaming and applications starting with ZEN5 so they habe to retest everything. He wrote, there is like a few specific situations where you can see a slightly lower score, but only if you know it. It's basically in the run to run variance region. So there's no 10 slower or sosomething.

Interesting. Would you have a link? I didn't find anything on Computerbase site.