Discussion Beyond zen 6

[adroc_thurston]

Fat brain NVL failing to beat X3D Zen5 would imply vanilla NVL being slower than 270K Plus in games. Imagine the fallout.

Not slower, but not much faster either.

 

[luro]

If the market won't bear the price (and for many CPU markets that is the case), raised costs simply lead to less profit, or in Intel's case, a loss.

AMD is only in a position to raise prices where there is no competing product in that segments consumer's eyes AND they are willing to pay the higher price for what they perceive is valuable. I would argue that in many markets AMD is still seen as an alternative to Intel, not a segment leader.

Yes, in DIY, gaming, and even in higher end DC, AMD is the recognized leader, but none of these markets are high volume (relatively speaking).

You seem to be part of the "sky is the limit" CPU pricing thinkers. The top of the line Zen 5 was introduced at $600. Zen 4 was introduced at $700.

I expect Zen 6 flag ship to come in around $600 as well.

Intel's flag ship launches for the last 2 generations have been ~ $590.

What is your expert opinion of where Zen 6 flagship will be launched at?

You have to consider X3D versions. AMD is already more expensive.
While Intel’s flagship has MSRP of $589 the 9950X3D costs $699 or $899 if you consider the X3D2.

The high volume part from AMD sits at almost $500, while Intel is at what? $350?

And I would expect at least a $70-100 price bump on every tier from both vendors.

 

[Win2012R2]

Ppl should prepare for a Zen 6% style of disappointment first.

3D version should be above 6 ghz, which will make it a very nice upgrade to those with Zen 4 3Ds:
1) 20%+ extra clock
2) 25%+ IPC from 2 gens
3) 50% more cores on same chiplet.

Combined: 225% faster than 8 cores Zen 4 3D.

Clocks might be even higher for a few cores - which is typical scenario for games with a few main threads needing it.

Yeah, that's worth $600, can be offset by selling older CPU for at least $250.

Just need a new BFG GPU for it.

 

[marees]

Zen 7 consumer 72c 🤔🤔🧐🧐

text summary

AMD Zen 7 Leak Reveals Major IPC Gains And Huge Cache Upgrades

Assuming the information is correct, AMD's upcoming Zen 7 processor architecture looks to be heavily focused on AI workloads.

hothardware.com

 

[OneEng2]

From the twit:
"UBS noted that Intel's foundry business is seeing improving prospects, particularly in the 14nm process."

I propose a toast:
To Intel 14nm - the cause of, and solution to, all of Intel's problems.

Lol. I think their problems run a bit deeper than that.

12c X3D will be $600 if NVL bLLC isn't completely awful.
24c well... $899 is kinda being telegraphed.

24c X3D for $999, take it or leave it.

Yea... about this, let's just put a pin in this and discuss again after the announcement.

AMD CCDs have been around 80mm2 lately. Hard to imagine them jumping to 100mm2 in the current desktop market, but certainly, dare to dream 😁.

 

[Doug S]

... but the premise of the leak (zen 7 launching on A14 in 2028) depends on that being the case.

Well sure if it ends up that TSMC begins A14 mass production in H1 2028 then it would be possible for AMD to "launch" Zen 7 by the end of the year. They don't need to have 20+ million chips installed in finished products on launch day like Apple does, and "launch" doesn't necessarily mean the same thing for them since they are selling components not finished products.

Anyway how often are rumors predicting release timing of products 2 1/2+ years out actually proven to be correct?

 

[adroc_thurston]

They don't need to have 20+ million chips installed in finished products on launch day like Apple does

Yeah they do.
It's a server CPU fiesta now.

Well sure if it ends up that TSMC begins A14 mass production in H1 2028

No, HVM means HVM.
Q3'28 HVM means products on shelves right that Q.

 

[Doug S]

Yeah they do.
It's a server CPU fiesta now.

No, HVM means HVM.
Q3'28 HVM means products on shelves right that Q.

No HVM means start of HVM. Finished chips don't appear out of the other end of the line an hour after HVM commences.

TSMC has ALWAYS talked about production in reference to its START, and every process requires more steps which lengthens the time between start of HVM and first finished chips appearing. Everyone is also using more advanced packaging which requires more steps and adds more time.

 

[adroc_thurston]

No HVM means start of HVM. Finished chips don't appear out of the other end of the line an hour after HVM commences.

TSMC has ALWAYS talked about production in reference to its START, and every process requires more steps which lengthens the time between start of HVM and first finished chips appearing. Everyone is also using more advanced packaging which requires more steps and adds more time.

man you really read into foundry PR too hard. get real.

 

[OneEng2]

You have to consider X3D versions. AMD is already more expensive.
While Intel’s flagship has MSRP of $589 the 9950X3D costs $699 or $899 if you consider the X3D2.

The high volume part from AMD sits at almost $500, while Intel is at what? $350?

And I would expect at least a $70-100 price bump on every tier from both vendors.

First, 9950X3D and X3D2 are VERY low volume parts. I think the only reason AMD bothered to do it AT ALL was because it didn't require much work to do it, and it gave them a little more breathing room while Intel did its refresh.

Intel's "High Volume" is NOT the same thing as AMD's "High Volume" pretty much need to multiply AMD HV * 10.... and High volume is 5 series, not 7 (ie ~$200).

One could now argue that Intel is offering more processing value than AMD in the high volume market. Not a great way to make a profit though.

AMD is eating up the lower volume, higher margin markets ..... especially DC. This makes a big difference.

I have no doubt that Z7 DC version will be a BEAST to behold. I fully expect it to finish off Intel in DC (at least religate them to the lower profit area for a decade IMO).

 

[adroc_thurston]

I can probably taze a pack of bonobos into understanding the concept of "winner takes all" before this guy gets it.

 

[luro]

9950X3D and X3D2 are VERY low volume parts

You are all over the place. I didn’t said it was. I was clearly comparing flagship CPUs prices from both. Where we a have a difference of $300 because AMD’s one is just better.

Intel's "High Volume" is NOT the same thing as AMD's "High Volume"

Yeah, Intel’s high volume is cheap CPUs with paper thin margins in mobile and oem/prebuilts. But I was specifically talking about DIY where AMD is dominating rn.
Whether you want or not AMDs high volume in this segment are the X3D versions. You can check the data online.

 

[Doug S]

man you really read into foundry PR too hard. get real.

Your source that TSMC is for the first time ever with A14 talking about manufacturing dates in terms of when finished chips exit the foundry?

Wait, I know this one! "Trust me bro" 🙄

 

[DrMrLordX]

I can probably taze a pack of bonobos

That might get you in trouble with some wildlife officials. But hey please document the process if you succeed!

 

[adroc_thurston]

That might get you in trouble with some wildlife officials

That's ok you can bribe them.

Your source that TSMC is for the first time ever with A14 talking about manufacturing dates in terms of when finished chips exit the foundry?

Wait, I know this one! "Trust me bro" 🙄

Please stop reading into foundry PR so hard.

 

[Mopetar]

Yeah they do.
It's a server CPU fiesta now.

What stops AMD from auctioning off whatever limited capacity they can manufacture to the highest bidder that will take what's available?

To some extent they already do this and the rest of the market only gets access months later when the hyperscalers are sated or production ramps to exceed their demand.

Right now it's even better to do that. What's the point of having a shiny new consumer CPU available if RAM costs just as much if not more than the CPU?

 

[adroc_thurston]

What stops AMD from auctioning off whatever limited capacity they can manufacture to the highest bidder that will take what's available?

It's not very limited and stuff like CSP inventory digestion prevents you from shipping infinite Turin at once.
Those servers have to go somewhere and power on somehow, you know.

What's the point of having a shiny new consumer CPU available if RAM costs just as much if not more than the CPU?

DIY you mean? Plugs into an existing socket with existing memory, but with higher perf and ASPs.
It's a good deal for everyone involved.

 

[dangerman1337]

I mean if Zen 7 ends up HVM in Q2 2028 if A14 production starts, then it's very doable to do Q3 2028 launch, I mean smaller CCDs won't be hard to manufacture if A14 ends up hitting targets. We're not talking about big 750mm2 dies here when everything else is like 6nm IODs or 4nm V-Cache.