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

[DisEnchantment]

Speculate at will

 

[RnR_au]

Unless you count N3B but eh whatever, Apple is getting all of that anyway.

Only for this year I thought.

 

[Joe NYC]

The ai boom will never be over. You're over simplifying what AI is if you only follow llms. That's amateur stuff on the scale of what's being done behind doors.

I meant in the current form where multiple companies believe they need to own a supercomputer sized AI hardware capability.

 

[Ajay]

Only for this year I thought.

Apparently for two years. Probably to maintain A17 production and for M3 SoCs for the Mac lineup.

 

[gdansk]

Only for this year I thought.

Who else is using N3B? It's almost moot.

 

[A///]

I meant in the current form where multiple companies believe they need to own a supercomputer sized AI hardware capability.

Oh well, you see Jensen has the ego the size of the sun, our galaxy, and beyond. Let him live his end years making nvidia into a giant.

 

[RnR_au]

Who else is using N3B? It's almost moot.

Apologies for not clarififying. Reports from earlier states that Apple had booked all of TSMC's N3 capacity for 2023.

 

[DrMrLordX]

Only for this year I thought.

Who else is using N3B? It's almost moot.

More to the point, is anyone going to bother with N3B except for legacy support after N3E becomes available?

 

[Doug S]

More to the point, is anyone going to bother with N3B except for legacy support after N3E becomes available?

Perhaps TSMC is not letting anyone use N3B without a commitment to moving to N3E when it comes online. If rumors are true that Apple will be switching A17s early next year in the middle of iPhone 15 Pro/Max production, there's no way they would let whatever other users of the process there may (cryptominers or whatever) continue to use it.

Why would they want to continue to support a process if there are no major customers usng it. You don't keep it online for a 1000 wpm or whatever the non Apple usage might be. And since we haven't heard confirmation of anyone else using it, it is possible Apple may be the ONLY customer for N3B.

 

[adroc_thurston]

it is possible Apple may be the ONLY customer for N3B.

Intel.

 

[Joe NYC]

Why would they want to continue to support a process if there are no major customers usng it. You don't keep it online for a 1000 wpm or whatever the non Apple usage might be. And since we haven't heard confirmation of anyone else using it, it is possible Apple may be the ONLY customer for N3B.

It seems that way. I am not aware of any other customer.

 

[Joe NYC]

Intel.

For which product?

 

[A///]

For which product?

ARL? It's been said for a while.

 

[adroc_thurston]

For which product?

Pretty much every ARL (68, 816).

 

[A///]

(68, 816).

?

 

[Joe NYC]

?

That would be 6P + 8E and 8P + 16E

 

[adroc_thurston]

?

6+8 mobile and 8 + 16 desktop ARL tiles are N3b.

 

[A///]

That would be 6P + 8E and 8P + 16E

Thanks. I've seen him post those numbers before and couldn't figure out what he was talking about.

I'm half convinced he's a robot.

 

[H433x0n]

Intel.

It’s the N3B variant?

 

[adroc_thurston]

It’s the N3B variant?

Yeah they've commited themselves to it eons ago.

 

[H433x0n]

Yeah they've commited themselves to it eons ago.

I knew they were using N3- I just didn’t know it was that specific variant. That really sucks that it’s not design rule compatible with N3E.

 

[adroc_thurston]

That really sucks that it’s not design rule compatible with N3E.

Well, N3e was never ever supposed to exist initially.
Things happen.

 

[Joe NYC]

Pretty much every ARL (68, 816).

That's quite interesting. What about a year ago+ looked like a coup by Gelsinger, to secure TSMC N3 capacity, during shortage, leaving AMD in the cold, now starts to look more and more like an unfortunate misstep.

I wonder if Intel is also paying also only for good die, like Apple is rumored to.

 

[A///]

I wonder if Intel is also paying also only for good die, like Apple is rumored to.

Yeah and I've got you a sea front property in Idaho to sell you.

 

[H433x0n]

That's quite interesting. What about a year ago+ looked like a coup by Gelsinger, to secure TSMC N3 capacity, during shortage, leaving AMD in the cold, now starts to look more and more like an unfortunate misstep.

I wonder if Intel is also paying also only for good die, like Apple is rumored to.

There’s absolutely zero shot they’re paying for known good die. They’re paying a ton per wafer for a node that’s honestly not great for their purposes.

Their decision to outsource really couldn’t have had worse timing. I would argue Zen 5 being on N4P is a much better situation from a cost / performance standpoint.

 

[A///]

There’s absolutely zero shot they’re paying for known good die. They’re paying a ton per wafer for a node that’s honestly not great for their purposes.

Their decision to outsource really couldn’t have had worse timing. I would argue Zen 5 being on N4P is a much better situation from a cost / performance standpoint.

Think of it this way, Haxxon. Get enough highly polished wafers with very little to no known good dies, place it in front of Gelsinger outside on the roof of Intel's building in santa clara, and he may just see AMD in the background.

Like Gelsinger said, AMD in the rear view mirror. Not the way he intended to.