Question Zen 6 Speculation Thread

[Fjodor2001]

If they include X3D already at Zen6 launch, what will they use as mid-life kicker?

 

[Thibsie]

Necessity

Doesn’t look like it, if Zen6 is as good as some say it is.
Anyway, they can still speed the x3d introduction if needs arises e.g. Zen5x3d.

 

[Win2012R2]

If they include X3D already at Zen6 launch, what will they use as mid-life kicker?

Better clocks on N2P? A tad lower prices? Free 32GB kit with each CPU?

As for the real shills, reddit is already echoing with voices claiming AMD does not support their GPUs.

Zen5 launch only really took off when X3D was announced - people will wait it out, and nobody on 9800X3D will upgrade because there is little chance bog standard Zen6 will be much quicker, only X3D will do.

 

[jdubs03]

If they include X3D already at Zen6 launch, what will they use as mid-life kicker?

Simple refresh akin to the 9850X3D.

A tad lower prices? Free 32GB kit with each CPU?

Lul.
——————
I would expect more folks upgrading from <= Zen 4, than from Zen 5. Typical of most product upgrade cycles. N-1 should see the least upgrades, aside from the enthusiasts.

 

[Win2012R2]

Lul.

Yeah I know...

I would expect more folks upgrading from <= Zen 4

Yeah for sure, I've got 7800X3D (well 16 cores, but 8 disabled to deal with scheduling stuff) - but I won't upgrade until X3D version is available and given what happened with Zen5 - short time to introduce it, then I'll certainly sit it out. At the very least AMD should at least announce it to ship in say 3 months later, this would be ok.

 

[Fjodor2001]

Better clocks on N2P? A tad lower prices? Free 32GB kit with each CPU?

Simple refresh akin to the 9850X3D.

I don’t think that’s in the same league as adding X3D. Except perhaps for the 32 GB RAM if the shortage persists. 🤑

 

[Thunder 57]

Yeah I know...

Yeah for sure, I've got 7800X3D (well 16 cores, but 8 disabled to deal with scheduling stuff) - but I won't upgrade until X3D version is available and given what happened with Zen5 - short time to introduce it, then I'll certainly sit it out. At the very least AMD should at least announce it to ship in say 3 months later, this would be ok.

And "Osborne effect" themselves? That would be stupid.

 

[Win2012R2]

And "Osborne effect" themselves? That would be stupid.

If they announce X3D at the same time as Zen6 is announced, with only catch that it will take say 2 months longer for release, there won't be Osborne effect - not saying nothing will make people disappointed as there is no way normal Zen6 will be much quicker than 9800X3D, and anybody on 7800X3D would certainly want X3D model.

 

[CakeMonster]

If they will dominate Intel at release, they're in no hurry with X3D, and it will also make the mid-generation refresh easier. They can just launch at a very good efficiency point with plenty of power and OC headroom, and then adjust when needed for the refresh.

 

[Thunder 57]

If they announce X3D at the same time as Zen6 is announced, with only catch that it will take say 2 months longer for release, there won't be Osborne effect - not saying nothing will make people disappointed as there is no way normal Zen6 will be much quicker than 9800X3D, and anybody on 7800X3D would certainly want X3D model.

They having nothing to gain by doing that. Likely nothing to lose either. People with a 7800X3D will wait whether they know the release date or not.

 

[Win2012R2]

People with a 7800X3D will wait whether they know the release date or not

Yes they will, but it will cause unnecessary agro, bad reviews (why bother with new Zen 6 when you can have 9800X3D?) and dumpen initial sales, just like it happened with Zen5 - so that's 99% chance AMD will do exactly that.

 

[StefanR5R]

Re big.LITTLE (meaning different core microarchitectures combined in a single CPU):

• Right now a common assumption is that the new LP cores will indeed differ microarchitecturally from the other cores.

• In the Zen 5 generation, AMD have cores with full-rate AVX512 and cores with half-rate AVX512, but they do not make CPUs in which both of these types appear. To me, it seems likely that they maintain the split of full-rate/ half-rate AVX512 equipped cores in the Zen 6 generation too. But unlike the tweet referenced in #9,102 implied, AMD might still not combine both types in a single CPU (except via LP cores).

 

[gdansk]

Doesn’t look like it, if Zen6 is as good as some say it is.

I'm pretty sure it isn't.

 

[LightningZ71]

If they announce X3D at the same time as Zen6 is announced, with only catch that it will take say 2 months longer for release, there won't be Osborne effect - not saying nothing will make people disappointed as there is no way normal Zen6 will be much quicker than 9800X3D, and anybody on 7800X3D would certainly want X3D model.

In general, that's probably a fare statement with respect to 1T. Here's the issue, that's going to be HIGHLY application dependent. There are some apps that will happily live in 48MB of L3 that weren't happy at all in 32MB that will do surprisingly well on Zen6. I expect that Zen6x3d will have the same characteristics that Zen5x3d had, in that accessing the 3d cache took a few cycles more than the first 32MB. On gaming benchmarks where people are splitting hairs over the difference between 148fps and 152fps, that latency difference will be evident.

Then you move on to the games that didn't fit into the 128MB 3d cache neatly, but do better with 144MB like Zen6 has, then move on to the games that have such ugly memory access patterns that the 3d cache just doesn't help much, then the better IoD link and likely improved performance of the memory controllers will help.

And what of situations where 8 cores becomes a limiting factor? Having 12 cores in the same segment will improve things, especially for games that like to do a lot of background shader work.

There are lots of reasons why a full single CCD 12 core CCD 3d cache chip might be notably faster than the 9850x3d, and I haven't even touched on the potentially higher core clocks and improved IPC of the cores themselves that are hotly debated here.

 

[basix]

I totally agree with that. Vanilla Zen 6 without V-Cache has the potential of exceeding 9800X3D performance in many games. But there are for sure cases where it is nearly impossible without having enough L3$ like e.g. "Anno 1800" (48MB vs. 96MB is still a 2x difference).

 

[ToTTenTranz]

Vanilla Zen 6 without V-Cache has the potential of exceeding 9800X3D performance in many games.

CPUs without V-Cache are probably going to be very dependent on RAM clocks, and fast RAM for the foreseeable future is even more ridiculously expensive than regularly clocked RAM.


I'd expect V-Cache Zen5 with 6 GT/s or slower memory to be faster than Zen 6 with the same memory.

 

[basix]

I'd expect V-Cache Zen5 with 6 GT/s or slower memory to be faster than Zen 6 with the same memory.

It really depends on Zen 6 latency reductions in general.

• Improved branch prediction
• Bigger OoO window
• Bigger INT register file
• Maybe bigger L1i$
• +50% L3$
• Advanced packaging
• N3 vs. N6 IOD (faster fabric clocks?) --> 1:1:1 FCLK / UCLK / MCLK

I would assume that Zen 6 has much better memory latencies compared to Zen 5, even with identical memory clock rates. And with the in-core updates you get additional benefits (IPC).
It will not be enough for all games but on average? Maybe.

 

[adroc_thurston]

CPUs without V-Cache are probably going to be very dependent on RAM clocks,

No, they're dependent on the fatness of the core.

 

[techjunkie123]

No, they're dependent on the fatness of the core.

I thought the core isn't getting much fatter. So minimal improvements outside of clocks?

 

[adroc_thurston]

I thought the core isn't getting much fatter

Indeed. It's still teens IPC.

So minimal improvements outside of clocks?

It's a normal tock.

 

[basix]

Indeed. It's still teens IPC.

For latency sensitive workloads it could be disproportionately more IPC compared to number crunchers like rendering.

 

[adroc_thurston]

For latency sensitive workloads it could be disproportionately more IPC compared to number crunchers like rendering.

Depends on the scope of front-end changes.

 

[LightningZ71]

In the end, RAM speed and latency changes are going to be Incremental at best. As much as the various cache levels hide ram performance, each change to ram performance is minimized.

 

[adroc_thurston]

In the end, RAM speed and latency changes are going to be Incremental at best

Ah, no, DDR5 straps are going all the way to 8000.
But memlat is incremental yes.

 

[Fjodor2001]

Ah, no, DDR5 straps are going all the way to 8000.
But memlat is incremental yes.

But what max speed will Zen6 support?