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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

And so there remains no incentive for AMD to actually produce it in volume.

stxH already outshipped their wildest expectations.
Also repair what OEM relationships lmao

 

[gdansk]

stxH already outshipped their wildest expectations.
Also repair what OEM relationships lmao

It's merely a rumor. But it seems believable.

As for sales, I'm sure it has since expectations for escaped lab subjects is often near nil. But it still doesn't give them any reason to make the gamer SKU at prices gamers can somewhat justify.

 

[DrMrLordX]

No, laptops are absolutely a market Strix Halo should be in.

Why? They can ship in volume to other markets. All OEMs want to do is slap on a 5070 or 5060 anyway and call it a day. Strix Halo takes a lot of configuration options away from OEMs, and OEMs don't like that. Joke's on them, Intel/NV will probably do the same thing pretty soon . . .

And without laptop designs there remains no incentive for AMD to actually produce it in volume.

Ummm

stxH already outshipped their wildest expectations.

Yeah, especially since it was just a proof-of-concept for later parts, no?

 

[Joe NYC]

Why? They can ship in volume to other markets. All OEMs want to do is slap on a 5070 or 5060 anyway and call it a day. Strix Halo takes a lot of configuration options away from OEMs, and OEMs don't like that. Joke's on them, Intel/NV will probably do the same thing pretty soon . . .

The real volume is in laptops.

The worst part is that the customers don't even have a chance to select one (with Strix Halo) over the alternatives, because OEMs are not offering them. Not giving consumers the choice.

But, there is a lot of "interest" in Nvidia version of Strix Halo. So, this is not exactly the problem of Strix Halo, but of OEMs or AMD standing with the OEMs.

Or more precisely Intel / NVidia orders to OEMs as @Jan Olšan speculates above.

 

[jpiniero]

But, there is a lot of "interest" in Nvidia version of Strix Halo. So, this is not exactly the problem of Strix Halo, but of OEMs or AMD standing with the OEMs.

That's because of AI, and nVidia = AI.

If the Halo products are going to stick around, there needs to be an actual market for local AI. Not just OEMs trying to push AI because of Wall Street.

 

[adroc_thurston]

especially since it was just a proof-of-concept for later parts, no?

It's just a new part in a new swimlane.

 

[Joe NYC]

It's just a new part in a new swimlane.

If AMD already reached their original target with mostly MiniPC sales of Strix Halo with one hand tied behind its back, then, offering 8 core + full GPU SKUs could significantly increase the demand.

But AMD still needs a lot more laptop models on the market, so there is a real choice between Strix Halo and NVidia N1X laptop.

It seems ridiculous there will be several X of Arm models in this "swimlane" vs. x86 models

 

[adroc_thurston]

But AMD still needs a lot more laptop models on the market, so there is a real choice between Strix Halo and NVidia N1X laptop.

There is no choice because the NV part does not work.

 

[jpiniero]

There is no choice because the NV part does not work.

Doesn't have to work if OEMs can win AI Hype points by releasing products with DGX.

 

[LightningZ71]

But that's just it, it underperforms AT PRESENT for its AI target market, and from the testing we've seen online, it also isn't very good at games YET and requires a LOT of software manipulation to do what it manages. AMD's product is performing as expected for AI, and stands to improve as AMD's software stack matures. It ALSO does quite well at gaming with MINIMAL effort. In addition, it's scalable enough to have secondary markets in handhelds and tablets.

We're it not for NVIDIA having more resources than most countries, DGX would have perished long ago.

 

[adroc_thurston]

Doesn't have to work if OEMs can win AI Hype points by releasing products with DGX.

It physically struggles to work with eDP laptop panels.

 

[DrMrLordX]

The real volume is in laptops.

Okay but AMD wasn't chasing volume with Strix Halo. They were chasing margins and opening up a new

It's just a new part in a new swimlane.

^^^ what he said.

If AMD already reached their original target with mostly MiniPC sales of Strix Halo with one hand tied behind its back, then, offering 8 core + full GPU SKUs could significantly increase the demand.

They actually did that (to my surprise), it's the AI Max+ 388:

https://videocardz.com/newz/amd-to-launch-gaming-oriented-ryzen-ai-max-388-strix-halo-apu-with-full-radeon-8060s-graphics

 

[LightningZ71]

And that volume MIGHT be sufficient to reduce the COGS enough to make it a viable product...

 

[Joe NYC]

And that volume MIGHT be sufficient to reduce the COGS enough to make it a viable product...

I wonder how laptop OEMs react to AMD finally offering a rational SKU for the product line, if AMD manages to get another laptop design or two.

And probably there will be an increase in MiniPC sales with this new SKU. I might be interested in one, but I don't need an upgrade yet...

 

[poke01]

I wonder how laptop OEMs react to AMD finally offering a rational SKU for the product line, if AMD manages to get another laptop design or two.

And probably there will be an increase in MiniPC sales with this new SKU. I might be interested in one, but I don't need an upgrade yet...

I really like the Halo line up, just waiting for RDNA5 halo then it will be my console for the living room.

 

[Abwx]

AMD chose not to partner with any big laptop OEM to assist with developing a laptop main board to base a family of laptop design on, unlike Intel has forever, and not in the way NVidia supports integration of their mobile graphics products.

AMD needs to learn that this isn't Field Of Dreams, if you just build it (the processor), and then do nothing else, they just won't come.

Minuscule manufacturers like GPD have the means to design several in house MBs for their handhelds and you re telling me that big OEMs with roughly 50bn revenue
need some help to do so..?..

Nowadays designing a proper MB is a cakewalk with modern CAD, otherwise there wouldnt
be those huge amount of different laptops, so obviously that s more of a willfull choice
to not use STH.

 

[Josh128]

RE: All the Strix Halo discussion, DF did a video about a month ago using the GMKtec EVO-X2 395 Max+ mini PC compared to both Nvidia and AMD desktop builds, all using a 9800X3D, as well as a vanilla PS5. Pretty interesting, TLDW results as follows:

Vs Nvidia: Sits between 4060 and 5060 (closer to 4060 though), stomps 3060, beats 3060 even in RT (and is fairly close to 4060)

Vs AMD: Equal to 5% faster than 6700 non-XT, moreso faster than 7600, and stomps 6600. Beats 6700 by 20%+ in RT.

Vs PS5: Essentially identical performance, with better power consumption (130W in 85W mode, 180W in 120W mode, PS5 220W). PS5 has double the VRAM bandwidth vs Strix Halo.

 

[LightningZ71]

Minuscule manufacturers like GPD have the means to design several in house MBs for their handhelds and you re telling me that big OEMs with roughly 50bn revenue
need some help to do so..?..

Nowadays designing a proper MB is a cakewalk with modern CAD, otherwise there wouldnt
be those huge amount of different laptops, so obviously that s more of a willfull choice
to not use STH.

Yes, they are mostly all able to do it on their own, IF THEY WANT TO. It is NOT free. They won't do it on their own if they don't think that they can command a price point sufficient to recover their investment at their targeted production volume. Intel and Nvidia mitigate much of this by doing a lot of the development work for these OEMs, so all they really need to do is put puzzle pieces together and book production after a few prototypes. AMD, as far as any of us have ever heard about, barely does this.

 

[ToTTenTranz]

RE: All the Strix Halo discussion, DF did a video about a month ago using the GMKtec EVO-X2 395 Max+ mini PC compared to both Nvidia and AMD desktop builds, all using a 9800X3D, as well as a vanilla PS5. Pretty interesting, TLDW results as follows:

Vs Nvidia: Sits between 4060 and 5060 (closer to 4060 though), stomps 3060, even in RT (and is fairly close to 4060)

Vs AMD: Equal to 5% faster than 6700 non-XT, moreso faster than 7600, and stomps 6600. Beats 6700 by 20%+ in RT.

Vs PS5: Essentially identical performance, with better power consumption (130W in 85W mode, 180W in 120W mode, PS5 220W). PS5 has double the VRAM bandwidth vs Strix Halo.

Remember those are measurements taken at the wall. When using a DC battery the power consumption should be significantly lower, over 20%.

 

[Josh128]

Remember those are measurements taken at the wall. When using a DC battery the power consumption should be significantly lower, over 20%.

Dont think the EVO-X2 has battery option, so moot point. Hypothetically though, the same should be true about the PS5.

 

[Abwx]

Yes, they are mostly all able to do it on their own, IF THEY WANT TO. It is NOT free. They won't do it on their own if they don't think that they can command a price point sufficient to recover their investment at their targeted production volume.

It cost peanuts and is largely recovered with all the rest of the laptop design xit something like 10-20k units, actualy it cost way more not doing it...

 

[jpiniero]

It cost peanuts and is largely recovered with all the rest of the laptop design xit something like 10-20k units, actualy it cost way more not doing it...

The Strix Halo handhelds are probally doing 10-20k sales combined.

 

[adroc_thurston]

The Strix Halo handhelds are probally doing 10-20k sales combined.

It's a part where every unit sold matters.

 

[Hulk]

I have some questions regarding extrapolating clockspeed.

If Zen 5 runs at 5GHz, 200W, all cores loaded on a particular workload running 16/32, what clockspeed would a theoretical 24/48 Zen 5 on the same process node at 200W, running the same (perfectly MT scaling) workload run?

Core count is increased by 50% or a factor of 1.5. Assuming linear scaling that would give us a low estimate of 3.33GHz, correct? In reality, might that number be ~3.7GHz?

Assuming a 20% efficiency boost on the new node might we see 4.44GHz at 200W for Zen 6 for a similar workload?

 

[eek2121]

Lmao people are delusional about getting an Strix HALO device for cheap

Also unrealistic. These parts aren't priced with inflated margins or anything. The vanilla RTX 4060, for example, was $300+. An 8 core Zen 5 processor for another $300+, and another $100-$300 for memory (prices are going up! so expect this part to also not be cheap), and you have an easy $1,000 for all of the above. Compare it with Zen 5 + RTX 5060 and you'll see the prices really aren't that far off. The prices of a Ryzen 9 270 with a 5060 and 32gb of RAM are $1,400+. The 8060S may be slower than the 5060, however between packaging, DRAM speeds, quad channel memory, and AI folks gobbling it up, I think things are actually normal from a price standpoint.

Fast, modern tech ain't cheap. Those are the rules of the game, unfortunately.

Side note: The top part is still on my shopping list. A mini PC running Linux/Steam sitting under my TV would be delightful, however that purchase is still a ways off.