Discussion Beyond zen 6

[LightningZ71]

You're talking rationally. MS is talking money. They have modeled that a dedicated NPU will be more efficient for CoPilot always on. It's not a "high performance" situation.

 

[Fjodor2001]

MS is pushing a subscription sales model. They wanted an always on AI component in Windows that you were willing to pay for. If it drained your battery rapidly or worked like dog excrement, no one would buy it. The NPU helps with both.

Not sure what you mean by always on AI component. Always on as in always available, yes. Always on as in only for always on AI functionality, no, since it'll be used for more that always on AI functionality.

W.r.t. NPU vs (i)GPU to provide necessary TOPS, does it matter as long as the TOPS are provided? If the NPU can provide the necessary TOPS using better TOPS/W and lower cost, I guess NPU is better option than (i)GPU.

 

[soresu]

MS is pushing a subscription sales model. They wanted an always on AI component in Windows that you were willing to pay for. If it drained your battery rapidly or worked like dog excrement, no one would buy it. The NPU helps with both.

IMHO the vast majority aren't going to buy based on AI regardless of performance.

All these companies are desperately trying to push it to justify the hardware build out and running costs, but the returns are never going to match that spending.

 

[LightningZ71]

Oh, I don't dispute either of your points. The NPU was supposed to provide the most efficient implementation then possible for accomplishing the needed tasks without blowing the budget and nuking battery life. They were trying to FORCE the market instead of MEETING the market and the result if a very unwanted copilot in windows 11.

 

[MadRat]

By zen 8 I fully suspect Quantum processing and AI to merge into a co-processor so we can make endless jokes about how itsso fast that not only did zen 8 kill the quantum cat in the box but it ate the AI. It will literally be so fast that when you decide to game it will already played it for you. Whoa, zen 8 might just end gaming.

 

[biostud]

Well if zen7 is also AM5, then I don't have to pay too much interest in Zen6, as I then will first do an upgrade when zen8 is launched.

 

[soresu]

Well if zen7 is also AM5, then I don't have to pay too much interest in Zen6, as I then will first do an upgrade when zen8 is launched.

It's the opposite for me as long as Zen7 is 32 core on the desktop.

That would mean a minimum 33% MT gain from Zen6 (assuming 24C SKU DT) even before IPC or clock frequency gains - which is a pretty decent gain for a single generation when you don't have to pay for a new mobo or RAM on top of the CPU.

 

[Makaveli]

It's the opposite for me as long as Zen7 is 32 core on the desktop.

That would mean a minimum 33% MT gain from Zen6 (assuming 24C SKU DT) even before IPC or clock frequency gains - which is a pretty decent gain for a single generation when you don't have to pay for a new mobo or RAM on top of the CPU.

good plan.

I'm just waiting for the 12 core single CCD X3D Zen 6 chip. If we see a 16 core Single CCD chip in Zen 7 that would be nice also. I don't need that many cores for my workloads.

 

[itsmydamnation]

getting zen6 x3d for me will depend on what GPU's look like. 7800X3D with a 7900XTX , i dont see the point of upgrading CPU that has ~ 40% more performance as 7900XTX will just be a massive bottleneck.

 

[Makaveli]

getting zen6 x3d for me will depend on what GPU's look like. 7800X3D with a 7900XTX , i dont see the point of upgrading CPU that has ~ 40% more performance as 7900XTX will just be a massive bottleneck.

I'm on a XTX also the plan is to go RDNA 5 later that year.

 

[marees]

I'm on a XTX also the plan is to go RDNA 5 later that year.

probably 2028 onwards

 

[Saylick]

getting zen6 x3d for me will depend on what GPU's look like. 7800X3D with a 7900XTX , i dont see the point of upgrading CPU that has ~ 40% more performance as 7900XTX will just be a massive bottleneck.

Just play at 1080p. You’ll see more gainz with Zen 6 X3D then. Problem solved.