Question Zen 6 Speculation Thread

[LightningZ71]

I wonder if they will increase the downstream link bandwidth to the chipset? 4 x PCIe 4.0 is a bit slow at the moment. 4 x PCIe 6.0 would keep pin count low while quadrupling bandwidth and would be manageable as it's all inside the board. If they could then give us 4-8 more exposed lanes downstream, doesn't even have to be PCIe 5.0 either.

 

[MS_AT]

You're getting 1-hi with 6.5.
Enjoy

Any updates about the AVX512? Will the implementation be the same in Zen6 desktop as in Zen5 desktop? Or will it go back to Zen4 scheme?

 

[511]

Any updates about the AVX512? Will the implementation be the same in Zen6 desktop as in Zen5 desktop? Or will it go back to Zen4 scheme?

Why would the regress lmao

 

[LightningZ71]

Well, at one point, the rumor mill had great divergence between consumer CCDs and server CCDs, to the point that all mobile and mainstream desktop parts were going to have the cut down mobile Zen5 core AVX-512 implementation and only Threadripper and server would have the full thing.

 

[Win2012R2]

all mobile and mainstream desktop parts

"desktop" chiplets are server parts, in mobile they already use Zen4 double pumping which is appropriate.

 

[marees]

"desktop" chiplets are server parts, in mobile they already use Zen4 double pumping which is appropriate.

That was the rumour that desktop will share mobile rather than server

 

[Win2012R2]

That was the rumour that desktop will share mobile rather than server

That would be mobile part re-purposed to be used in desktops, the only reason server parts won't be used in desktops is if say Intel fails and AMD will need more server parts, though in this case I reckon they will have other bottlenecks than chiplets, but who knows, big money can solve lots of problems quick.

 

[yottabit]

I don't think I will be able to resist the 12 core Zen 6 at 7 GHz with 2 layers of V-Cache. 😉

Don’t forget the SMT4!

 

[StefanR5R]

the only reason server parts won't be used in desktops is if [...]

Actually, already since Zen 5, desktop CCDs are no longer used in servers (IOW they come from separate manufacturing batches; they are no longer merely different bins), apparently because the frequency target of desktop caused AMD to tune the desktop CCD transistors way outside server comfort zone.

Any updates about the AVX512? Will the implementation be the same in Zen6 desktop as in Zen5 desktop?

My guess: Yes. Reasons: N2P; and Strix Halo precedence.
Edit: And engineering can explain it to bean counters by mentioning "AI".

 

[Win2012R2]

Actually, already since Zen 5, desktop CCDs are no longer used in servers

It seems to me that this is just parametric binning, unless CCDs have different masks those are the same chiplets

 

[Kaffeekenan]

You're getting 1-hi with 6.5.
Enjoy

Did you just confirm the clock speeds of Zen 6 or was this a jokepost? Because I don't think even AMD knows the final specs yet.

 

[Josh128]

Did you just confirm the clock speeds of Zen 6 or was this a jokepost? Because I don't think even AMD knows the final specs yet.

Nobody knows yet. Nobody.

 

[Win2012R2]

Nobody knows yet. Nobody.

Apart from MLID...

 

[511]

Apart from MLID...

how can MLID be wrong he is the absolute source of truth in the world. Truth stems from MLID. Whatever MLID says is set in stone

 

[StefanR5R]

It seems to me that this is just parametric binning,

For what it's worth, the stepping denominations differ too, apparently. I admit that I have only seen 3rd parties saying that these are made differently, and that this is counter to AMD's modus operandi until Zen 4 inclusive. {edit: see next post}

[Also keep in mind that even Strix Halo, regardless of its low volume, has got its very own special CCD which is not interchangeable with Granite Ridge or Turin. This is in contrast to Zen 4 MI300A and MI300C which (according to AMD) have the same CCD as Genoa and Raphael despite their different GMI physical implementation.]

 

[adroc_thurston]

Actually, already since Zen 5, desktop CCDs are no longer used in servers

since Zen3.

and that this is counter to AMD's modus operandi until Zen 4 inclusive.

Server has lived in its own land for a while.

 

[Jan Olšan]

since Zen3.

Is there any proof they aren't made from the same masks as desktop ones?

 

[adroc_thurston]

Is there any proof they aren't made from the same masks as desktop ones?

Yes but you need to pay for TechInsights die analysis for that pmuch.

 

[StefanR5R]

Is there any proof they aren't made from the same masks as desktop ones?

Not an actual proof but it coincides: The steppings have different denominations.

 

[Timmah!]

how can MLID be wrong he is the absolute source of truth in the world. Truth stems from MLID. Whatever MLID says is set in stone

He knew before AMD did :-D

 

[Kryohi]

He knew before AMD did :-D

Well of course, AMD sets as internal frequency targets whatever MLID says first.

 

[Win2012R2]

MLID is God.

You've heard it on here first...

 

[511]

MLID is God.

You've heard it on here first...

MLID is beyond God he has evolved passed Godhood.

 

[igor_kavinski]

Oh, mlid just knows how to finger mess with the Matrix

 

[yottabit]

Guys I had a great dream last night

Lisa Su told me we'll get:

• Medusa Halo with on-package memory
• Zen 6 Dual CCD X3D V-cache with MR-DIMM support
• RDNA4 Discrete GPU with 96 GB VRAM for < $3,000

I wish I never woke up