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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

So no new info about Zen5 from AMD at CES 2024.

Why would they osbore their current stuff?

Do you still think it’ll be released and available in stores in April 2024, like some on this forum have said previously?

yeah

 

[Fjodor2001]

Why would they osbore their current stuff?

They wouldn't necessarily have to. Depends on what info they would release. But at least some teaser would have been nice.

 

[adroc_thurston]

They wouldn't necessarily have to.

Oh but it will.

 

[Fjodor2001]

Regardless of what info it is? Why?

And does it mean you think AMD will not provide any info at all about Zen5 until the same day that it's released and available in stores?

 

[adroc_thurston]

Why

It makes everything before it obsolete, instantly.

And does it mean you think AMD will not provide any info at all about Zen5 until the same day that they are released and available in stores?

Pretty much!

 

[jpiniero]

I'd say it's more likely that since Wall Street is obsessed with AI, they aren't going to talk much about client products that don't have NPUs. The Zen 5 APUs are probally months away from release..

 

[adroc_thurston]

I'd say it's more likely that since Wall Street is obsessed with AI, they aren't going to talk much about client products that don't have NPUs. The Zen 5 APUs are probally months away from release..

No, it's all about not osborning current parts.

 

[CakeMonster]

I was looking at RAM for a possible Z5 build later this year, and wondering why we're at the max of 48GB per stick (96GB kits). You'd think 64GB sticks (128GB kits) would be possible if 48GB is. Is that a limitation of market demand, current chipsets, CPU's, DIMM physical space, or what? Also interesting that there's currently way fewer EXPO kits than XMP at 96GB.

 

[adroc_thurston]

I was looking at RAM for a possible Z5 build later this year, and wondering why we're at the max of 48GB per stick (96GB kits).

Because 24Gb ICs is what ships.

Is that a limitation of market demand, current chipsets, CPU's, DIMM physical space, or what?

You can only drive so many ranks without register buffering.

 

[gdansk]

Samsung started production on 32 gigabit DDR5 chips late last year. Micron was sampling in late 2023 and plan to start mass production early 2024. These should allow affordable 64GB DDR5 UDIMM.

 

[CakeMonster]

Thanks for that info, will register buffering be an issue when 32Gb IC is available (whenever that might be), or is that a problem for even higher capacities ( > 64GBx2 kits)?

 

[adroc_thurston]

will register buffering be an issue when 32Gb IC is available

No.

or is that a problem for even higher capacities ( > 64GBx2 kits)?

Yes that's for RDIMMs aka things not present on client platforms.

 

[igor_kavinski]

You'd think 64GB sticks (128GB kits) would be possible if 48GB is.

Motherboard support is already there!

ASRock > News

www.asrock.com

 

[Fjodor2001]

It makes everything before it obsolete, instantly.

If it turns out that Zen5 will only bring ~20% IPC increase, same max 16C as Zen4, slight frequency regression, no new IO-die & iGPU, while still at 170W TDP, and $999 for top SKU, there's no osbourning to be concerned about. If anything, I'd say it would be a reverse osbourning to inform about it at this point.

 

[adroc_thurston]

If it turns out that Zen5 will only bring ~20% IPC increase, same max 16C as Zen4, slight frequency regression, no new IO-die & iGPU, while still at 170W TDP, and $999 for top SKU, there's no osbourning to be concerned about.

I wish, but alas, it's the single biggest jump in CPU perf since ~2006 (outside of phones, that is).

 

[gdansk]

None of that impacts the actual release date. Only when they disclose it.
In any case I, Nostradamus, doubt any announcement of desktop parts before Computex.

 

[adroc_thurston]

doubt any announcement of desktop parts before Computex.

Computex is for Strix1 (and RDNA4).

 

[Fjodor2001]

I wish, but alas, it's the single biggest jump in CPU perf since ~2006 (outside of phones, that is).

Zen1->Zen2 in 2019 went from 8C to 16C, roughly doubling the CPU perf. So ~20% increase for Zen4->Zen5 would be nothing in comparison. If we'll see 32C on Zen5 AM5 (or some combination of P + E cores reaching similar MT performance core count) then it would be something similarly impressive. Otherwise, at a substantial price increase to $999 for top SKU, Zen5 will be a reverse osbourne to announce info about at this point.

 

[adroc_thurston]

Zen1->Zen2 in 2019 went from 8C to 16C, roughly doubling the CPU perf. So ~20% increase for Zen4->Zen5 would be nothing in comparison.

See you're thinking there's no free lunch.
But there is.

If we'll see 32C on Zen5 AM5 (or some combination of P + E cores reaching similar MT performance core count) then it would be something similarly impressive

You need faster cores, not more of them.
Let's just quote an eons old article.

 

[H433x0n]

So no new info about Zen5 from AMD at CES 2024.

Do you still think it’ll be released and available in stores in April 2024, like some on this forum have said previously?

My guess is that the launch plans are similar to Zen 4. There's an event by the end of April that goes over performance and pricing followed by general availability a month later.

Zen 4 had a launch event 8/29/2022 with public release on 9/27/2022.

 

[Fjodor2001]

You need faster cores, not more of them.

Need better MT perf. Also need substantially better perf/watt (more than IPC increase can provide, so E cores would be needed) and ideally lower TDP. If Zen5 will only be 16C and have specs and price as previously mentioned, it’ll be a reverse osbourne for AMD to inform about it now.

 

[adroc_thurston]

Need better MT perf

Ugh, no?
again, I've quoted the article, you only have to find it.

Also need substantially better perf/watt

It's >20% iso-ish node so gucci.

If Zen5 will only be 16C and have specs and price as previously mentioned

Again, read.

 

[Fjodor2001]

Ugh, no?
again, I've quoted the article, you only have to find it.

Maybe for your workloads ST perf is only what matters, not for mine. I need MT perf too.

It's >20% iso-ish node so gucci.

Then core count increase is needed, possibly introducing E cores to get better perf/watt. If price is bumped to $999 I’m expecting way more. Otherwise the discounted Zen4:s will be a much better option.

 

[adroc_thurston]

Maybe for your workloads ST perf is only what matters, not for mine.

all workloads need 1t.

Then core count increase is needed

No.

Just read it.

Otherwise the discounted Zen4:s will be a much better option.

It's gonna be instantly obsolete.

 

[gdansk]

Need better MT perf. Also need substantially better perf/watt (more than IPC increase can provide, so E cores would be needed) and ideally lower TDP. If Zen5 will only be 16C and have specs and price as previously mentioned, it’ll be a reverse osbourne for AMD to inform about it now.

I really disagree. 1T is what everyone will benefit from. AMD needs to improve that the most. You say 20% IPC with perhaps a 5% frequency reduction? That's average. Maybe the cheapest X3D part would sell well but that's it.

But what I don't get is how there is harm in announcing a product that is average like that? You would want to announce it early. You do it before your competition. It'd look good enough compared to Raptor Lake but not Intel's subsequent products (unless they suck too but you must not count on that). Not buying the reverse Osbourne idea.