Question Zen 6 Speculation Thread

[adroc_thurston]

if it is N2P it is launching H2 27.

I think you're mistaking N2p for 18A-P. Happens.

 

[Tigerick]

My take is Zen6 with 12-core and 48MB L3 cache will be fabbed by N3P and will launch by Q4-2026; maybe slightly early..

Zen 6c with 32-core and 128MB L3 cache will be fabbed by N2。

Zen6 will be used for Client and Server (96-core or lower). There won't have any high clock Zen6 die with more than 96-core due to power constraints and memory bandwidth limitation. Unless AMD decides to raise the TDP to more than 680 - 840W, the only option to buy HP ST server CPU is Turin 128-core.

I will bookmark this page and revisit once AMD announce Zen 6 / 6c.

 

[511]

I think you're mistaking N2p for 18A-P. Happens.

You are the one that is mistaken not me 🤣

 

[Tigerick]

I think you're mistaking N2p for 18A-P. Happens.

Geez, you should at least read the TSMC's transcript before make a mistake correcting someone:



Got it?

 

[eek2121]

Geez, you should at least read the TSMC's transcript before make a mistake correcting someone:

View attachment 123709

Got it?

The user he quoted claimed N2P wasn’t launching until 2027. @adroc_thurston was pointing out this is false.

 

[Tigerick]

The user he quoted claimed N2P wasn’t launching until 2027. @adroc_thurston was pointing out this is false.

Geez, second half HVM means the product will be shipping by 2027 is correct...

 

[511]

The user he quoted claimed N2P wasn’t launching until 2027. @adroc_thurston was pointing out this is false.

Look at the patten N2 is H2 25 HVM and the product would ship in H2 26 and N2P is H2 26 HVM so it would be at least H2 27 Product shipping to consumer.

 

[Joe NYC]

Look at the patten N2 is H2 25 HVM and the product would ship in H2 26 and N2P is H2 26 HVM so it would be at least H2 27 Product shipping.

No, that's a different node called N2PA - A for Aged. It works like with aged steaks, where you take a perfectly fine steak, but "age" it for 30-60 days and then sell it for more money.

In case of N4PA node, you take a wafer with perfectly fine chips, then set it aside for a year, and only then sell it (for higher price). It is aimed at discerning customers, who don't mind paying extra and are in no hurry.

 

[adroc_thurston]

Look at the patten N2 is H2 25 HVM and the product would ship in H2 26 and N2P is H2 26 HVM so it would be at least H2 27 Product shipping to consumer.

Indeed, NVL-S is now an H2'27 product.
Sorry, H1'28. Forgot to account for Intel time.

 

[511]

Indeed, NVL-S is now an H2'27 product.
Sorry, H1'28. Forgot to account for Intel time.

NVL is N2 not N2P 🤣 if we account for Intel time than do it properly it is +++++ years so NVL is 2031 Product such a shame people don't know how accurate Intel is with their '+'.

 

[adroc_thurston]

NVL is N2 not N2P

Wrong.

 

[511]

Wrong.

Oh really lol? They already Tape Out the A0 on N2 and 18AP.

 

[Kaffeekenan]

Indeed, NVL-S is now an H2'27 product.
Sorry, H1'28. Forgot to account for Intel time.

How certain are you? I thought NVL was 2026.

 

[adroc_thurston]

How certain are you?

It's a joke.

They already Tape Out the A0 on N2

Everyone does.
Then, then things happen.

 

[soresu]

Sorry, H1'28. Forgot to account for Intel time.

Mmmhmmm, Intel seem to be executing on CPUs even worse than AMD have been on gaming GPUs.

(RDNA4 was hella late, overstock RDNA3 or not)

 

[adroc_thurston]

Mmmhmmm, Intel seem to be executing on CPUs even worse than AMD have been on gaming GPUs.

That's a joke.

RDNA4 was hella late

It's like 1Q late.

overstock RDNA3 or not

They were just waiting until FSR4 is ready.

 

[soresu]

It's like 1Q late.

Relative to previous gen -> gen timelines, not whatever arbitrary internal roadmap they have.

Considering how far behind nVidia they have been in RT (forget DLSS I couldn't care less about it) they should have just rebranded the RDNA3 hw as lower end 9xxx and shouldered the cost.

 

[Thunder 57]

Relative to previous gen -> gen timelines, not whatever arbitrary internal roadmap they have.

Considering how far behind nVidia they have been in RT (forget DLSS I couldn't care less about it) they should have just rebranded the RDNA3 hw as lower end 9xxx and shouldered the cost.

Then they'd get crapped on as "Rebrandeon". No win. Besides you can't go on gen on gen all the time that is why there are roadmaps. Sometimes products are given less priority or there is a problem during design/development. In this case it seems FSR4 wasn't ready. Between releasing it with FSR broken and getting a bad name and delaying it a couple months and getting it right the choice seems obvious. I'm sure AMD isn't losing sleep over you not liking that decision.

 

[soresu]

THen they'd get crapped on as "Rebrandeon". No win

Exactly - they can't win so they might as well do it anyway given nVidia does the same thing.

 

[soresu]

In this case it seems FSR4 wasn't ready. Between releasing it with FSR broken and getting a bad name and delaying it a couple months and getting it right the choice seems obvious

Doesn't follow that they needed to release it with FSR4 at all.

FSR2/3 is inferior to DLSS2/3 for sure - but not broken.

The only legitimate reason for the delay in my mind is building up adequate stock that it doesn't become scalper bait automatically, which I can understand them being extremely careful about overdoing, considering the problems they already have with RDNA3 in the channels.

Also given how laughably bad nVidia's RTX 5xxx launch has been thus far I really don't think that it would have made that much of a difference had AMD launched FSR4 later than RDNA4.

You might say that AMD couldn't have known that, but nVidia do seem to have been following a trend of diminishing returns in each new gen in the last few years coupled with their famously cheap stance on RAM, so it's not like the worst parts of this launch were totally unpredictable.

I'd say that nVidia are gonna have to dig deep to win gamers back, but I think at this point either they no longer care because they believe AI revenue matters more, or they just believe their brand is so embedded in the customer base that they will continue to 'upgrade' to these minimal hardware improvements (in raster anyways) despite the price hikes.

 

[inquiss]

Doesn't follow that they needed to release it with FSR4 at all.

FSR2/3 is inferior to DLSS2/3 for sure - but not broken.

The only legitimate reason for the delay in my mind is building up adequate stock that it doesn't become scalper bait automatically, which I can understand them being extremely careful about overdoing, considering the problems they already have with RDNA3 in the channels.

Also given how laughably bad nVidia's RTX 5xxx launch has been thus far I really don't think that it would have made that much of a difference had AMD launched FSR4 later than RDNA4.

You might say that AMD couldn't have known that, but nVidia do seem to have been following a trend of diminishing returns in each new gen in the last few years coupled with their famously cheap stance on RAM, so it's not like the worst parts of this launch were totally unpredictable.

I'd say that nVidia are gonna have to dig deep to win gamers back, but I think at this point either they no longer care because they believe AI revenue matters more, or they just believe their brand is so embedded in the customer base that they will continue to 'upgrade' to these minimal hardware improvements (in raster anyways) despite the price hikes.

I don't agree with any of your takes in this or the previous few posts. The launch went pretty well. I expect they're happy.

Even if I disagree with your suggestions of what they should have done, I'm going to again remind another poster (you in this case), that AMD or any other company can't base past decisions on our knowledge of the present. That's the future to those lot from the past, and they didn't know what was going to happen.

 

[soresu]

I'm going to again remind another poster (you in this case), that AMD or any other company can't base past decisions on our knowledge of the present

I explained my reasoning on this very clearly.

It didn't require foreknowledge of the future.

Only some bare minimum reasoning of nVidia's business priorities regarding the dGPU market having shifted significantly from 5 years ago, priorities advertised with great frequency from every possible quarter of late.

The problems in nVidia's current launch have precedence in business practices of previous launch years from which to draw reasonable conclusions that do not require the Delphic oracle to resolve from thin air.

 

[inquiss]

I explained my reasoning on this very clearly.

It didn't require foreknowledge of the future.

Only some bare minimum reasoning of nVidia's business priorities regarding the dGPU market having shifted significantly from 5 years ago, priorities advertised with great frequency from every possible quarter of late.

The problems in nVidia's current launch have precedence in business practices of previous launch years from which to draw reasonable conclusions that do not require the Delphic oracle to resolve from thin air.

Saying you explained it, and then immediately reasoning that AMD should have known the future... if only they could have reasoned it the way you did, it was so obvious...with your current knowledge...of the present.

Blackwells specs are very decent. You'd expect more performance, something went wrong. AMD could not have known they'd produce a dud, or not produce enough etc. They couldn't... unless of course they could see the future, like you.

 

[adroc_thurston]

Considering how far behind nVidia they have been in RT

They just do it different.

and shouldered the cost.

the cost of?

 

[Win2012R2]

reasoning that AMD should have known the future...

AMD employs people who are paid very big $$$ to predict the future, every big company got 'em.

AMD could not have known they'd produce a dud, or not produce enough etc. They couldn't...

They should have bloody counted on it! Sooner or later Nvidia was bound to slip up and focus on AI was clearly indicating they are very likely to neglect consumer side.