Discussion AMD SoC Halo series GPU discussion

[fastandfurious6]

Its good Medusa Halos is still a thing. The rumors of quitting that avenue after one shot

what rumours? who made such nonsense rumours 🤣🤣 halo is literally the future it just makes sense

 

[igor_kavinski]

halo is literally the future it just makes sense

Not near future. Medusa Halo will get released after consumer Zen 6 and probably even after 10950X3D.

 

[jpiniero]

what rumours? who made such nonsense rumours 🤣🤣 halo is literally the future it just makes sense

Main issue is whether OEMs are really interested in Halo, esp if AI hype is dying.

 

[LightningZ71]

That bubble is popping soon. We're already seeing some roadmaps get torn up with follow on consolidation. It's not going to immediately flatline though. There are political winds blowing that will lead capitalists to favor AI over restoring certain positions.

 

[fastandfurious6]

the next era of computing will be defined when fridges have local GPT-4

LLMs will continue even if hype somehow dies

 

[poke01]

the next era of computing will be defined when fridges have local GPT-4

I would rather not. I don’t my fridge to hallucinate.

 

[branch_suggestion]

https://twitter.com/x/status/1959631046871548052

CES 2028 gonna hit like a cement truck.

 

[gdansk]

Hmm. Doesn't 3 years seem way too long for Strix Halo to be an elite experience?
Who's going to buy (basically) RDNA3 in 2027?

 

[adroc_thurston]

Doesn't 3 years seem way too long?

yeah, mdsH is Q2'27.

 

[branch_suggestion]

yeah, mdsH is Q2'27.

I sure hope so, it would be a really horrible gap otherwise.
Computex 2027 looking like the big one then.

 

[poke01]

yeah, mdsH is Q2'27.

I wonder if N1X will be out by then

 

[adroc_thurston]

I wonder if N1X will be out by then

Who tf cares, it's gonna be cursed like every other Tegra in existence.

 

[Joe NYC]

Main issue is whether OEMs are really interested in Halo, esp if AI hype is dying.

Some of the AI predictions say AI on client will arrive late in the cycle, in 2-5 years. But what do the really know?

 

[misuspita]

So Medusa Halo on that chart is....AWOL?

 

[marees]

So Medusa Halo on that chart is....AWOL?

Or 2028 🤔😉

 

[ToTTenTranz]

So Medusa Halo on that chart is....AWOL?

And so is Medusa Mini-Halo / Premium.

Honestly that roadmap sounds too bad to be true. Going for a full 3 year cadence for a Halo product is the best way to ensure you don't have a Halo product for most of the time.

 

[igor_kavinski]

I think they will decide once they are able to recoup the development costs for Strix Halo.

In the meantime, there's hope for a Strix Halo Refresh with faster RAM.

 

[ToTTenTranz]

I think they will decide once they are able to recoup the development costs for Strix Halo.

I.e. never, at the current rate.

 

[LightningZ71]

Well, you've got to actually offer them for sale if you want to actually sell them... Having less than a handful of actual shipping, purchasable "laptop" designs out there for what was ostensibly it's target market isn't going to do you volume any favors. Now, it does look like they'll move positively tens of micro-desktops in a variety of highly similar designs, so that should make an impact on numbers. Maybe not more than a pillow against the titanic, but an impact none the less.

 

[gdansk]

I think they will decide once they are able to recoup the development costs for Strix Halo.

With TSMC 5nm at capacity it's cheaper to ignore the sunken costs fallacy and not make more of a product that doesn't sell well. So until that problem is "resolved", no cheap big dies for you.

 

[ToTTenTranz]

Now, it does look like they'll move positively tens of micro-desktops in a variety of highly similar designs, so that should make an impact on numbers. Maybe not more than a pillow against the titanic, but an impact none the less.

Looks like most of those designs are simply OEMs ordering small amounts of the same reference board.
Strix Halo was supposed to rule over high performance thin&light laptops. It didn't, though.

With TSMC 5nm at capacity it's cheaper to ignore the sunken costs fallacy and not make more of a product that doesn't sell well. So until that problem is "resolved", no cheap big dies for you.

It is?

 

[gdansk]

It is?

Reports point to yes. They could be or become wrong.