Question Zen 6 Speculation Thread

[poke01]

What who said that lmao

idk why there is so much so hype here and on twitter about APX, is it simply because Nova lake will have it and zen6 won't?

It doesn't matter for most end users at all unless software is complied for it.

 

[adroc_thurston]

But do you know if Zen 6 is better than 9%?

Yeah.
Not by much, though.

 

[511]

idk why there is so much so hype here and on twitter about APX, is it simply because Nova lake will have it and zen6 won't?

It doesn't matter for most end users at all unless software is complied for it.

Yup that's the issue though knowing Intel they will just introduce it in Geekbench or Maybe new Cinebench as well

 

[Jan Olšan]

idk why there is so much so hype here and on twitter about APX, is it simply because Nova lake will have it and zen6 won't?

It doesn't matter for most end users at all unless software is complied for it.

Well, from AMD fan point of view, new x86 gooddies are nice even if they come from Intel.
Particularly when you don't have to fear that Intel will manage to kill competition (or at least it looks like we don't have to ATM - they still have majority share in all the segments which show how much the market still sucks, although it's still better than the thing we have in GPU market).

 

[Joe NYC]

9% is not that much. But do you know if Zen 6 is better than 9%?

Zen 6 should have IPC (~10%) + clock speed increase (~10%-15%)

 

[adroc_thurston]

Zen 6 should have IPC (~10%) + clock speed increase (~10%-15%)

It has low teens IPC (think 13%) and whatever for fmax.
Probably 20%.

 

[gdansk]

I do not find 6900MHz plausible.
Though the ES 192 core parts having +300MHz on even the dense variants (where it is only a single node improvement) is promising.

 

[adroc_thurston]

I do not find 6900MHz plausible.

How is this different from Apple or QCOM pumping speeds

 

[511]

How is this different from Apple or QCOM pumping speeds

They are not pumping 6 GHz yet. QCOM 5 GHz SKU isn't even available that's even more vaporware than 5.1 Ghz PTL.

 

[Win2012R2]

It doesn't matter for most end users at all unless software is complied for it.

This is where iBOT comes in

 

[rainy]

My expectations are still the same: 10-15 percent higher IPC and similar range for clock.

 

[adroc_thurston]

They are not pumping 6 GHz yet.

Yes.
A14, they will.

QCOM 5 GHz SKU isn't even available that's even more vaporware than 5.1 Ghz PTL.

Do you understand how silly this sounds?
Pegasus is a solid chunk higher IPC you know.

 

[511]

Yes.
A14, they will.

Nice Cope

Do you understand how silly this sounds?
Pegasus is a solid chunk higher IPC you know.

while as silly it sounds it ain't shipping yet

 

[adroc_thurston]

Nice Cope

?
You're getting 5.5 on N2X already.
A14 is a few bins on top of that.

while as silly it sounds it ain't shipping yet

My point is that 2.8k and 4k .gb6 pts share the N3-class process node and have roughly similar v/f ranges.
Don't pretend to be dimwitted for once.

 

[511]

My point is that 2.8k and 4k .gb6 pts share the N3-class process node and have roughly similar v/f ranges.
Don't pretend to be dimwitted for once.

Nice one to create artificial delta 388H scores 3.0K ish for reference device and Snapdragon get like 6% boost from SME the v/f range is similar but my point remains they are not shipping yet which is a fact. It's you who is ignoring that.

?
You're getting 5.5 on N2X already.
A16 is a few bins on top of that.

N2X is Fmax optimized node TSMC themselves said its like 5% fmax over N2P which has fmax improvement over N2 small % add up we are comparing base A14.

 

[adroc_thurston]

Nice one to create artificial delta 388H scores 3.0K ish for reference device

With DDR5.
LP5x scores are 2850-ish. Same as STX1.

for reference device and Snapdragon get like 6% boost from SME the v/f range is similar but my point remains they are not shipping yet which is a fact. It's you who is ignoring that.

Okay, 3900, not 4k. Happy now?
But QCOM ships 4.74 bin in a phone and that one still claps PTL handily. You can buy it everywhere.

N2X is Fmax optimized node TSMC themselves said its like 5% fmax over N2P which has fmax improvement over N2 small % add up we are comparing base A14.

I'm talking product-level freq targets on IHV roadmaps, not vague foundry numbers based off ROs and friends.

 

[ToTTenTranz]

Xbox gets one too. But not Valve.

Sony, Microsoft, Intel and Zhongshan Subor get semicustom chips.

Valve got lots of money, it's them who is cheap here.

At this rate Deck 2 will be Intel+nGreedia

These two sentences are mutually exclusive. If Valve is cheap, they're not going with Nvidia.
I do wish Valve had hired Intel for Steam Deck 2, though. Panther Lake is awesome and the G3 line will probably be a generational upgrade over the Deck at its ~10W optimal TDP.

 

[adroc_thurston]

Sony, Microsoft, Intel and Zhongshan Subor get semicustom chips.

Eons old stuff.
AMD is a rich and busy company now.

I do wish Valve had hired Intel for Steam Deck 2, though. Panther Lake is awesome and the G3 line will probably be a generational upgrade over the Deck at its ~10W optimal TDP.

Intel gfx IP is terrible (in general, but especially at 10W) and ANV is a joke.
They were never ever considered for even half a second. It was always QCOM vs AMD.

 

[gdansk]

It's not even about being rich or lazy. Zen 6 has more configurations which cover more use cases while the cost of silicon tape out has risen.

 

[adroc_thurston]

Zen 6 has more configurations which cover more use cases while the cost of silicon tape out has risen.

Cost is whatever.
Valve can pay. But the people are fairly finite, and a full semi-custom part is somewhat manpower-intensive.

 

[ToTTenTranz]

Intel gfx IP is terrible (in general, but especially at 10W)

This is provably incorrect.










The X9 388H - designed for ~25W and without the optimizations coming to the G3 line, is between 20 and 50% faster than the Steam Deck. AMD's own Strix Point and Hawk Point perform much worse at 10W.

 

[adroc_thurston]

is between 20 and 50% faster than the Steam Deck

You do understand that the thingy there is like 2 nodes ahead?
I'm not even talking 50mm^2 of GFX area alone.

 

[Thibsie]

These two sentences are mutually exclusive. If Valve is cheap, they're not going with Nvidia.
I do wish Valve had hired Intel for Steam Deck 2, though. Panther Lake is awesome and the G3 line will probably be a generational upgrade over the Deck at its ~10W optimal TDP.

Tell that to Nintendo 🤭
Ok, bad example I know…

 

[luro]

This is provably incorrect.

View attachment 141325

View attachment 141327

View attachment 141329

View attachment 141328


The X9 388H - designed for ~25W and without the optimizations coming to the G3 line, is between 20 and 50% faster than the Steam Deck. AMD's own Strix Point and Hawk Point perform much worse at 10W.

I think it should be better since is 2nm vs 4nm (or 5nm)? Not impressive

 

[adroc_thurston]

I think it should be better since is 2nm vs 4nm (or 5nm)? Not impressive

It's N3-ish stuff versus the N7 VGH.