Question Zen 6 Speculation Thread

[Joe NYC]

Not happening.

Maybe what is happening is that the V-Cache die is designed so that there could be multiple layers, but the decision to turn it into a product is deferred for later, depending on market conditions.

 

[adroc_thurston]

Maybe what is happening is that the V-Cache die is designed so that there could be multiple layers

They already are.
Ever since the very original Zen3 V$.

 

[511]

I have a feeling the 52 core part is a dual 445K part but with better bins.

I have a question to the folks knowledgeable with Nova lake, will it also double the iGPU, Media engines and I/O ie like the M3 Ultra OR is it just the CPU title getting doubled?

The iGPU is Xe3/Xe3P based Media engine has been upgraded to Xe4 with Xe3 display engines integrated TB5 4 extra PCIE Lames everything is getting improved.
Looks like I did read the question correctly it's just one more compute tile.

 

[adroc_thurston]

I have a question to the folks knowledgeable with Nova lake, will it also double the iGPU, Media engines and I/O ie like the M3 Ultra OR is it just the CPU title getting doubled?

It's just a response to AMD 2t parts.
If you want a chungus gfx, opt for NVL-AX.

 

[511]

It's just a response to AMD 2t parts.
If you want a chungus gfx, opt for NVL-AX.

Btw it may have MoP to deal with signalling issues AMD faces with Strix Halo

 

[DrMrLordX]

Nnnope

Then it's gonna be an expensive part to hit their stated margin targets.

 

[igor_kavinski]

Btw it may have MoP to deal with signalling issues AMD faces with Strix Halo

Maybe not?

https://videocardz.com/newz/g-skill-demonstrates-ddr5-camm2-memory-at-10000-mt-s-on-asus-z890-motherboard

 

[Darkmont]

Maybe what is happening is that the V-Cache die is designed so that there could be multiple layers, but the decision to turn it into a product is deferred for later, depending on market conditions.

https://www.anandtech.com/show/16026/tsmc-teases-12-high-3d-stacked-silicon The technology AMD uses to 3D stack L3, SOIC, has been shown to support 12-hi stacks since 2019 with test chips. It shouldn't come as a surprise that AMD's got the ability in 2025/2026 to go above 1-hi layers

 

[adroc_thurston]

Maybe not?

https://videocardz.com/newz/g-skill-demonstrates-ddr5-camm2-memory-at-10000-mt-s-on-asus-z890-motherboard

gamer stable overclocks != 256b MoB LPDDR

Then it's gonna be an expensive part to hit their stated margin targets.

Targets are aspirational, they're never hitting them in this comp env.

 

[branch_suggestion]

I'm gonna make an argument now for 12c Z6 X3D pricing.

Firstly, the current 12 core X3D part launched at $599, a terrible price of course to upsell the 16c.
It is also a good bit weaker in gaming than the 9800X3D which launched at $479.
The new 12c part will not only be a lot stronger in gaming, but productivity too over the existing 12c part.
Now what about Intel? Well if the bLLC is really only on the 2 tile flagship then Intel cannot compete in gaming perf on price, period.
And there really is a level of productivity performance where far fewer people start to desire more cores, which the 12c should have more than enough nT perf for nearly everyone in client.

Finally, since it is AM5, you can price higher as many customers will have an existing motherboard, so no need to worry about double dipping scaring people.
Still lots of AM4 people, the leap in core count may be enough for lots of them to jump ship.
As AMD can make a very convincing slideshow against both the 9800X3D and the bad 9900X3D with Intel being stuck in a ravine somewhere, I bet the price will be $599.

 

[Thunder 57]

Funny, cause the Zen6 24/48 core part will match or likely come very close to the 48/48 part.

Going by past information here, it won’t be as good at clocking as high as Zen6, IPC will be “good”, area of the P core will be bad relative to Zen6.

As for the IPC we don’t know but I wonder if going to 32 registers will help. Who knows

Not without recompiling. I like the idea of APX but I am not certain it will do much for most of us.

 

[Thunder 57]

I'm gonna make an argument now for 12c Z6 X3D pricing.

Firstly, the current 12 core X3D part launched at $599, a terrible price of course to upsell the 16c.
It is also a good bit weaker in gaming than the 9800X3D which launched at $479.
The new 12c part will not only be a lot stronger in gaming, but productivity too over the existing 12c part.
Now what about Intel? Well if the bLLC is really only on the 2 tile flagship then Intel cannot compete in gaming perf on price, period.
And there really is a level of productivity performance where far fewer people start to desire more cores, which the 12c should have more than enough nT perf for nearly everyone in client.

Finally, since it is AM5, you can price higher as many customers will have an existing motherboard, so no need to worry about double dipping scaring people.
Still lots of AM4 people, the leap in core count may be enough for lots of them to jump ship.
As AMD can make a very convincing slideshow against both the 9800X3D and the bad 9900X3D with Intel being stuck in a ravine somewhere, I bet the price will be $599.

We don't even know if there will be a 16C part. That would be highly inefficient. Two 12C CCD's to get 16C? They would be better off doing 18 or 20 as an intermediate core count.

 

[Darkmont]

Maybe not?

https://videocardz.com/newz/g-skill-demonstrates-ddr5-camm2-memory-at-10000-mt-s-on-asus-z890-motherboard

A 48 minute stress test using very likely binned components says nothing about the ability for a different module with different memory that has different signaling requirements to operate at 24/7 JEDEC speeds in 85C+ temperatures.

 

[branch_suggestion]

We don't even know if there will be a 16C part. That would be highly inefficient. Two 12C CCD's to get 16C? They would be better off doing 18 or 20 as an intermediate core count.

I was referring only to the 9950X3D.
18c would be 12+6 which is not how AMD does SKUs and 20c would be 10+10 instead of 12+8 because once again, AMD wants you to pay up.
A 6c part will be a late salvage R3 maybe, never sold in a 2 CCD part.
8c+8c maybe but that is a 1/3 disabled part, for an R5 sure along with a 10c R5.
$599 for 12c X3D, $699 for 20c X3D, $799 for 24c X3D would work out fine.

 

[Thunder 57]

I was referring only to the 9950X3D.
18c would be 12+6 which is not how AMD does SKUs and 20c would be 10+10 instead of 12+8 because once again, AMD wants you to pay up.
A 6c part will be a late salvage R3 maybe, never sold in a 2 CCD part.
8c+8c maybe but that is a 1/3 disabled part, for an R5 sure along with a 10c R5.
$599 for 12c X3D, $699 for 20c X3D, $799 for 24c X3D would work out fine.

AMD doesn't use uneven CCD's. So 18 is probably out. My mistake. I think the closest they ever came was two cores linked together on the 3100, whereas the 3300X had all four together. Something like the 9900X is 6+6, not 8+4. That would hurt performance in particular L3.

 

[511]

The parts would likely be a combination of 12/10/8 Cores
i.e. 8/10/12/16(8+8)/20(10+10)/24(12+12)

 

[ondma]

I was referring only to the 9950X3D.
18c would be 12+6 which is not how AMD does SKUs and 20c would be 10+10 instead of 12+8 because once again, AMD wants you to pay up.
A 6c part will be a late salvage R3 maybe, never sold in a 2 CCD part.
8c+8c maybe but that is a 1/3 disabled part, for an R5 sure along with a 10c R5.
$599 for 12c X3D, $699 for 20c X3D, $799 for 24c X3D would work out fine.

Actually, if the CCD is in fact 12 cores, they should just skip a 10 +10 core X3D. It is just a part without a real use case. Probably worse for gaming than either the 12 or 24 core X3D, and intermediate in nT performance at higher cost than the 12 core. If the 20 core X3D were to be 12 +8, it would be quite nice though. You would get a full 12 X3D cores for gaming and an extra 8 cores for nT use. It also depends on what clocks they can get out of these higher core count chips. So far, AMD has been able to get the highest clock speed on the highest core count chip. Lets see if they can continue that with these higher core counts.

 

[branch_suggestion]

Actually, if the CCD is in fact 12 cores, they should just skip a 10 +10 core X3D. It is just a part without a real use case.

This is true, but trying to sell off salvages for a premium to the ignorant is always going to be attempted.
Still a 10c salvage is way more viable than 6c.

 

[Win2012R2]

Dual layer 3D might come only if they do it for servers, which they might given that they skipped Turin 3D.

Not sure added latency will be worth it for games - 144 MB is a very nice increase already, it does not matter that core count is +50% - it's meant to be used for same game app running exclusively more or less, not +50% apps concurrently.

 

[Io Magnesso]

Not happening.

Well, right, 240MB is too much

 

[Josh128]

12c X3D is going to be an ultra popular choice. If they can manage to get boosts up to~5.7 or more, its going to be considered a GOAT chip. If they manage to get ~6GHz out of it,

 

[511]

12c X3D is going to be an ultra popular choice. If they can manage to get boosts up to~5.7 or more, its going to be considered a GOAT chip. If they manage to get ~6GHz out of it,

View attachment 126745

sorry to put your hopes down but Intel has a bLLC variant without having to limit themselves to X3D frequency limitations they would have other issue though

 

[Io Magnesso]

Well, right, 240MB is too much

Cache… Cash increases...

 

[yottabit]

If you wanted your competitor to panic, wouldn't you leak FAKE roadmaps that scare them more than the real plans would?

The big brain play is to leak the real roadmap, but make Intel think it’s a fake

 

[Win2012R2]

If they manage to get ~6GHz out of it,

Couple of cores running at that level for UE5 games will do the trick, don't need all core for that