Question Zen 6 Speculation Thread

[511]

You do you but there is a non zero chance for that so upto you

 

[Win2012R2]

I don't think it is likely that a good core gets damaged during packaging.

Seems most likely to me that the main yield issue with 3D packaging is that 3D cache won't work due to alignment etc, but that surely can be turned off and sold as normal chip or special OEM or what not. Can misalignment damage cores too? No idea, but intuitively feels unlikely.

Either way the volume is the answer - since 3D chips are not highest volume anyway then surely salvage should be even lower volume? Guess they can accumulate them over period of time and sell like right about now.

 

[511]

since 3D chips are not highest volume anyway then surely salvage should be even lower volume? Guess they can accumulate them over period of time and sell like right about now.

That's what I was trying to say 😅

 

[Doug S]

No one uses igpus to game. So it can be tiny. Tiny tiny tiny.

They need to be good enough to handle everything EXCEPT gaming and CAD. How big (in mm^2) is the GPU in AMD's 9x00 family? My 9600X's iGPU seems to be adequate for general desktop use, and if it wasn't I have the dGPU upgrade option. They need a lot more for mobile because OEMs want to get rid of dGPUs outside of the desktop replacement segment. They're the ones driving AMD/Intel to continue enchancing the iGPU in their mobile CPUs.

 

[adroc_thurston]

because OEMs want to get rid of dGPUs outside of the desktop replacement segment

That's bull.
They really, really, really don't.

 

[blackangus]

They need to be good enough to handle everything EXCEPT gaming and CAD.

Id modify that to: Gaming, CAD/Simulation, and AI.

 

[adroc_thurston]

and AI.

bubble's on its last legs man

 

[poke01]

bubble's on its last legs man

Gonna pop when China stops buying AI hardware from American companies

 

[adroc_thurston]

Gonna pop when China stops buying AI hardware from American companies

That's kind of not very relevant.
It's just that LLMs plateau'd, are operationally expensive while being generally unreliable.
Doesn't help that altman was promising hailmary-terminator-machinehitler and GPT-5 is meh.

 

[blackangus]

bubble's on its last legs man

Hopefully!
But that doesnt mean it wont stay around in a normalized fashion.

 

[blackangus]

It's just that LLMs plateau'd, are operationally expensive while being generally unreliable.

They do have a place for somethings but are certainly not the panacea that was talked about.

I had to laugh my ass of when I was told at work by someone on the other end of a process in relation to an item which I am directly responsible for: ChatGPT doesn't say that (My product area) exists, you must be wrong.
Sad part is they were serious.

 

[Darkmont]

bubble's on its last legs man

The real bubble was the billions we wasted along the way

 

[igor_kavinski]

The real bubble was the billions we wasted along the way

Wasted? NO. The money went to the bubble makers who will find new ways to create more bubbles. Worthwhile investment for the future of humanity with more temporary jobs and more mansions and islands for the billionaires. Some day they will move to Mars so we gotta be patient

 

[soresu]

Gonna pop when China stops buying AI hardware from American companies

Given their inability to use TSMC for domestic designs or buy new (even old now) ASML hw for their own domestic fabs they may well have little choice until their domestic fab tooling and EUV litho can catch up (assuming it can).

 

[MadRat]

AI will plateau. It can only scale so much in software until there is some future unrealized breakthrough. What will not plateau as fast is general computing. If you build it (better) they will come.

 

[Doug S]

When the guy leading the charge says AI is in a bubble then you have to take him seriously. Usually the insiders benefiting from a bubble are keen to hype it up and ride the wave as long as possible, but I think he knows that the bigger it inflates the bigger the bust will be. He doesn't want the bust and disillusionment that follows to be so bad it hurts OpenAI's future.

 

[Joe NYC]

I went to the mountain, and I was given these words of wisdom, carved on marble tablets, to share with you:

 

[511]

I went to the mountain, and I was given these words of wisdom, carved on marble tablets, to share with you:

View attachment 129070

While you went to mountain I am going to sea to extract some salt so I can digest what the wisdom you brought us means.

 

[511]

duplicate

 

[marees]

While you went to mountain I am going to supermarket to buy some salt so I can digest what the wisdom you brought

You can get salt in the mountain also. But it is a premium variety

 

[Joe NYC]

You can get salt in the mountain also. But it is a premium variety

Yup, salt that's so good it rocks.

 

[511]

Yup, salt that's so good it rocks.

Rock you mean getting turned into a rock by seeing Medusa(Point/Halo)

 

[igor_kavinski]

I went to the mountain, and I was given these words of wisdom, carved on marble tablets, to share with you

Dear Brave Soul,

Thank you.

 

[igor_kavinski]

I see he is spewing incredible BS because he can't help it, the cretin, this time saying that a dGPU could have 128GB VRAM.

How wonderful.

EDIT: Dear God, it goes up to 512GB VRAM.

I think I may not survive consuming the amount of salt required.

 

[Joe NYC]

I see he is spewing incredible BS because he can't help it, the cretin, this time saying that a dGPU could have 128GB VRAM.

How wonderful.

EDIT: Dear God, it goes up to 512GB VRAM.

I think I may not survive consuming the amount of salt required.

The low end to mid dGPUs would be using LPDDR instead of GDDR.

And a professional / AI version of the same card could put in the biggest LPDDR chips that is currently on the market.

I find it quite intriguing that AMD is able to contain so much of the bandwidth requirements using on die L2s, to the point that AMD can get away with LPDDR memory.