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

[DisEnchantment]

Speculate at will

 

[ToTTenTranz]

Says you?

Says the amount of design wins in laptops (1, one, uno).

 

[DrMrLordX]

You are lucky, Framework not only has it as one of the first sellers, they even sell it starting from $1099 for the computer or $799 for the contained mini ITX mainboard.

I really don't understand these complaints about pricing and availability while you got these offers by Framework.

Framework has the same price for Ryzen AI 395+ w/64GB as everyone else: $1599. There are actually a few pre-orders for 128GB models for around $1699, though one of the models reportedly has weirdly low GB6 MT scores so nobody knows what's up with that model. Until they start shipping, nobody really knows how that thing is gonna work out.

Oh and according to Framework, they've already sold out 10 batches of the $1599 model. You'll be waiting until Q3 to get part of batch 11.

Most likely, Strix Halo wasn't cheap until AMD failed miserably at getting design wins in the laptop space.

What are you talking about? It isn't cheap now! Strix Halo laptops are selling secondhand on eBay for $3k-$4k.

If they sell a version with single 8c CCD + full 40CU IOD for like $400 they're probably already getting a markup well over 50% for it.

They'd be leaving money on the table.

 

[adroc_thurston]

Says the amount of design wins in laptops (1, one, uno).

2, and it's a new swimlane.

 

[Joe NYC]

Says you?

If Lunar Lake is supposedly universally hated by OEMs and still have ~20x laptop design wins of Strix Halo, then it is hard to see any silver lining for Strix Halo in laptops.

I have not seen a single one in its ideal form factor of 16-17", which would make it a decent workstation.

The evidence (comparing Lunar Lake to Strix Halo) now shows OEMs are just fine with MoP and not having MoP greatly contributed to Strix Halo fail in laptop space.

 

[poke01]

The evidence (comparing Lunar Lake to Strix Halo) now shows OEMs are just fine with MoP and not having MoP greatly contributed to Strix Halo fail in laptop space

Not sure how MoP comes into the equation. Lunar Lake was popular because it’s a more mainstream product and much cheaper than Halo. halo is more targeted to AI folks.

 

[gdansk]

If Lunar Lake is supposedly universally hated by OEMs and still have ~20x laptop design wins of Strix Halo, then it is hard to see any silver lining for Strix Halo in laptops.

I have not seen a single one in its ideal form factor of 16-17", which would make it a decent workstation.

The evidence (comparing Lunar Lake to Strix Halo) now shows OEMs are just fine with MoP and not having MoP greatly contributed to Strix Halo fail in laptop space.

Strix Halo is questionable. Adding a medium-sized Radeon GPU to a Zen CPU isn't really adding much value for consumers since no one wants a Radeon GPU, gamers nor AIsloppers. Especially RDNA3.x. But it sure adds a lot of cost that AMD wants their high margin on.

 

[OneEng2]

Well, according to a prolific poster here, there's no way that thing can cost anything less than $2000+. They should sell dozens!

LOL. Not sure which poster you are referring to, but I would agree with them. There isn't much market for > $1K system. Rumor is that the PS Pro will be around $700 .... but that is a niche IMO. Today, the going rate for a gaming system for home use is around $500.

I would also wonder about the need for 16 cores for gaming of any kind. Better off spending a bit of money putting a big wad of 3D cache on it instead of adding another compute tile.

If Lunar Lake is supposedly universally hated by OEMs and still have ~20x laptop design wins of Strix Halo, then it is hard to see any silver lining for Strix Halo in laptops.

I think it is a niche market. I do wonder how cost effective the design work was though.

Then again, Lunar Lake on N3B is likely not profitable at all for Intel IMO. I believe Intel is currently purchasing market share with their life blood.

 

[ToTTenTranz]

2,

One laptop and one fugly tablet (not a laptop).

it's a new swimlane.

So was Llano, yet it got dozens of design wins at the first try despite bundling ancient CPU cores for mobile.
STX Halo is going into a dozen mini-PCs instead, which is where other failed solutions went to die.



Trying to paint STX Halo's current market penetration as a win is a really poor choice, at this point.

They'd be leaving money on the table.

LOL what money? OEMs aren't touching it with a 10 feet pole.
They're leaving money on the table with the current plan.

 

[Joe NYC]

Strix Halo is questionable. Adding a medium-sized Radeon GPU to a Zen CPU isn't really adding much value for consumers since no one wants a Radeon GPU, gamers nor AIsloppers. Especially RDNA3.x. But it sure adds a lot of cost that AMD wants their high margin on.

The biggest fail was with the OEMs, since there are zero consumer laptops.

Whoever came up with the idea that OEM will do all the work, with signal integrity of 256 bit data interface FOR AMD is smoking crack.

It should be obvious at AMD that for OEMs, AMD laptops are an afterthought.

They were busy with endless Meteor Lake, Raptor Lake, Lunar Lake, even freaking Qualcomm laptops before they get to AMD. And taking a look at the amount of work needed for Strix Halo, they just all said "Never mind" in unison.

 

[adroc_thurston]

One laptop and one fugly tablet (not a laptop).

That's laptops.

So was Llano, yet it got dozens of design wins at the first try despite bundling ancient CPU cores for mobile.

No, mobile parts existed long before Llano.

STX Halo is going into a dozen mini-PCs instead, which is where other failed solutions went to die.

That would be Very True if stxH had GT4e levels of design volume.
alas!

 

[DrMrLordX]

LOL what money? OEMs aren't touching it with a 10 feet pole.
They're leaving money on the table with the current plan.

Why are you so fixated on niche Strix Halo laptops when $1600+ Strix Halo mini PCs are going through presale right now and are apparently selling out?

 

[Darkmont]

Given that Intel will have NVL-AX, Nvidia's N1X, and ugh I guess whatever Qualcomm craps out with SDXE 2, it's clear the "fat iGPU fat LPDDR pile" AKA M4 Pro/Max windows edition swim lane is a growing one with clear demand. At least enough that Medusa Halo's a thing and Intel's investing in their own parts. AMD's making a bigger bet on the ongoing SOCification of everything since apple. Also, STX-H not having MoP is a nothingburger since you'd need 4-8 COTS packages to fill that bus without going to bespoke 8CH packages (192-bit ones coming with LPDDR6 per JEDEC). Strix Halo wasn't hurt by a lack of MoP it was hurt by not having enough supply for the copious demand AMD did not anticipate. Also, MoP is pretty much dead for Intel since OEMs don't like how it locks them into SKUs and they want to source the memory themselves. It's a good trick for -Y level power but it's $$$ and so Intel won't be doing that again. Unless you believe RZL-M would be MoP for ? reasons

 

[jpiniero]

Why are you so fixated on niche Strix Halo laptops when $1600+ Strix Halo mini PCs are going through presale right now and are apparently selling out?

Volume of the Framework devices is probably very tiny.

Given that Intel will have NVL-AX, Nvidia's N1X, and ugh I guess whatever Qualcomm craps out with SDXE 2, it's clear the "fat iGPU fat LPDDR pile

It's not "Fat IGPU", it's AI. AIAIAIAAIAIAIAIAIAIAIAIAIAAIAIAI

Ayy

Eye

 

[poke01]

Volume of the Framework devices is probably very tiny.

I haven't seen a single framework laptop in Australia. Its very niche

 

[poke01]

I haven't seen a single framework laptop in Australia. Its very niche

doesn't help if you have no retail presence. You need retail presence to make volume

 

[moinmoin]

Oh and according to Framework, they've already sold out 10 batches of the $1599 model. You'll be waiting until Q3 to get part of batch 11.

Just to note, all of the batches are Q3 (there is an "All Batches" link showing the whole list), with the first batch being "early Q3". And Q3 starts tomorrow.

 

[Hail The Brain Slug]

Volume of the Framework devices is probably very tiny.

HP listed the Z2 Mini G1a for sale last week and there were.... 5 units of most of the SKUs.

Definitely the opposite of volume.

 

[eek2121]

Dell, HP and Lenovo are ready for N1X.
No one sane will be doing AI work on a Qualcomm laptop.
Agree but they do compete in the laptop space. And Apple’s laptops do sell well in the $1800+ range.

It’s not easy, laptops are very hard. Don’t expect NV to flood the market. These aren’t dGPUs.

PC laptop makers just don’t want to spend $$$ to make an actual premium laptop. Believe it or not, a laptop can be light, silent, fast, and cool at the same time.

If Lunar Lake is supposedly universally hated by OEMs and still have ~20x laptop design wins of Strix Halo, then it is hard to see any silver lining for Strix Halo in laptops.

I have not seen a single one in its ideal form factor of 16-17", which would make it a decent workstation.

The evidence (comparing Lunar Lake to Strix Halo) now shows OEMs are just fine with MoP and not having MoP greatly contributed to Strix Halo fail in laptop space.

You don’t see it in laptops because it isn’t designed for laptops. It is a scalable AI solution. Some companies are shoving them into laptops, but only because they want to have “mobile AI” covered.

Where the part really shines is in mini PCs.

The only way you will get cheap STX halo is if the bubble pops.

 

[yottabit]

PC laptop makers just don’t want to spend $$$ to make an actual premium laptop. Believe it or not, a laptop can be light, silent, fast, and cool at the same time.

You don’t see it in laptops because it isn’t designed for laptops. It is a scalable AI solution. Some companies are shoving them into laptops, but only because they want to have “mobile AI” covered.

Where the part really shines is in mini PCs.

The only way you will get cheap STX halo is if the bubble pops.

I see people shilling pretty hard for Strix Halo as a "scalable AI solution" as you say but to me it's very underwhelming in that regard, ESPECIALLY for mini PC form factor where you can get a Mac Mini or Studio with much more memory bandwidth.

LLM inference is almost ALL membw... and sure, Halo is better than your average PC if you're trying to solve on CPU only but it's much worse membw than say a 3090.

If you're a home AI enthusiast looking to do serious inference you'd be much better off buying 2x 3090 used or something and throwing them in a potato computer with a decent amount of RAM than a Strix Halo mini PC. Having 48 GB in VRAM at 1 TB/s+ and 64 GB offloaded to system RAM is going to be faster than having 128 GB on Strix Halo almost all of the time, especially if you're talking about newer models with Mixture of Experts or something. I don't think it can even come close to competing with 1x 3090.

Or heck you could get a last gen Epyc system with 8 or 12 channels of memory and run models off CPU.

Where Halo has an undeniable value proposition is in Windows laptops for gaming or other professional 3D graphics workloads, where you care about battery life. I'd love to see a serious Macbook Pro 16" competitor, something with a giant battery beautiful screen and not much else.

With AI workloads I can't imagine anyone picking a Strix Halo-based system over a Mac, it's not like they will need Windows OS and probably the only reason they wouldn't pick a Mac is they need CUDA, in which case Halo isn't going to help.

So really it's the niche of "Home AI enthusiasts who care about power efficiency, don't need to run models super fast, need Windows for some reason but don't need CUDA"... it's niche of a niche of a niche

I'm also still so disappointed that Halo can't take advantage of the full membw for CPU-based workloads.

Mini PC for gaming, sure, it makes total sense for that and has basically zero competition.

EDIT: I forgot to add with enough 3090's or M- Max chips you can actually conceivably do model training... not something I'd want to attempt with Strix Halo.

And what exactly makes it "scalable"? Do any of the mini PC have 10Gbe or Thunderbolt 5 standard like the Mac Mini/Studio to allow clusters at high bandwidth and low latency?

 

[fastandfurious6]

AMD failed miserably at getting design wins in the laptop space

tell you what.

I remember seeing Lisa Su bringing on stage 3 VPs from lenovo, asus and other 2 big vendors (I think HP and dell? idk), one by one, about partnerships and new models et cetera

Lisa was genuinely all like "AMD is happy we have strong partnerships" but the VPs were kinda sour. One example, Lisa held a lenovo laptop together with lenovo VP it was supposed to show LENOVO + AMD on screen, it did for 1 millisecond and then it went black lol

then you see in reality lenovo has 1000 intel designs and 5 amd designs


the only VP who seemed genuine was from ASUS, 35yr old 100% Taiwanese born raised and owned company, which is in practice the only company with actual close partnership on laptop segment, for many years either heavily promotes AMD designs beginning from Zephyrus and the Duo laptop or has chip exclusivity like the 7945HX3D only found on asus Strix laptops, was the only laptop design with a 3Dcache chip for years until zen 5, and so on



So there's three possibilities:

1 - amd's laptop sales/pr/planning team sucks

2 - amd's laptop sales/pr/planning team does good job but the game is rigged, OEMs continue gutting AMD laptop designs for no clear reason

3 - all of the above


call me crazy if you want but all I saw when Gelsinger was literally dancing on stage with a chip on his hand, all I saw was desperation, it was a very sad sight
then after 1 month he gets fired, out of the blue


btw is the site broken? I made this post literally in this page 🤣

 

[Markfw]

the site was broken yesterday. All seems fine today

 

[maddie]

the site was broken yesterday. All seems fine today

btw is the site broken? I made this post literally in this page 🤣

Have either of you noticed a slowdown in serving pages the past few days? I wonder if the owners have downgraded the servers as a first step to shutdown.

 

[StefanR5R]

AMD won _all_ of the designs (that is, 100%) of x86 quadchannel notebooks. And did so at their first attempt. :-P


(Now, if some of you don't like that there aren't more of such designs, then AMD may be partly to blame for that, but only partly.)

 

[igor_kavinski]

I wonder if the owners have downgraded the servers as a first step to shutdown.

They better not dare unless they let all of us download our posts.

 

[marees]

tell you what.

I remember seeing Lisa Su bringing on stage 3 VPs from lenovo, asus and other 2 big vendors (I think HP and dell? idk), one by one, about partnerships and new models et cetera

Lisa was genuinely all like "AMD is happy we have strong partnerships" but the VPs were kinda sour. One example, Lisa held a lenovo laptop together with lenovo VP it was supposed to show LENOVO + AMD on screen, it did for 1 millisecond and then it went black lol

then you see in reality lenovo has 1000 intel designs and 5 amd designs


the only VP who seemed genuine was from ASUS, 35yr old 100% Taiwanese born raised and owned company, which is in practice the only company with actual close partnership on laptop segment, for many years either heavily promotes AMD designs beginning from Zephyrus and the Duo laptop or has chip exclusivity like the 7945HX3D only found on asus Strix laptops, was the only laptop design with a 3Dcache chip for years until zen 5, and so on



So there's three possibilities:

1 - amd's laptop sales/pr/planning team sucks

2 - amd's laptop sales/pr/planning team does good job but the game is rigged, OEMs continue gutting AMD laptop designs for no clear reason

3 - all of the above


call me crazy if you want but all I saw when Gelsinger was literally dancing on stage with a chip on his hand, all I saw was desperation, it was a very sad sight
then after 1 month he gets fired, out of the blue


btw is the site broken? I made this post literally in this page 🤣

View attachment 126424

My guess. Something broke the site

So they rolled back to previous stable version