Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[Gideon]

Hmm, it seems AMD actually made its first 4-socket HPC server for MS with only CPUs and monstrous memory bandwidth available. 88x Zen 4 cores clocked at 4Ghz. Four of these have access to 450GB of HBM at ~7.0Tbps bandwidth. Nodes connected via InfiniBand.

Announcing Azure HBv5 Virtual Machines: A Breakthrough in Memory Bandwidth for HPC | Microsoft Community Hub

Discover the new Azure HBv5 Virtual Machines, unveiled at Microsoft Ignite, designed for high-performance computing applications. With up to 7 TB/s of memory...

techcommunity.microsoft.com

Microsoft Ignite 2024 Book of News

The Book of News is your guide to all key items being announced at Microsoft Ignite 2024.

news.microsoft.com

Microsoft convinced AMD and Nvidia to build a CPU with extraordinary features but it will never go on sale: 4th gen 9V64H has 96 cores and uses InfiniBand technology

Build exclusively for Microsoft Azure, this processor is set to deliver jaw-dropping performance

www.techradar.com

I guess under the hood it's just a MI300 derivate with only CPU cores? (Sort of like MI300A but no GPU cores, binned to 88 cores, down from 96)

 

[LightningZ71]

There's a Genoa HBM thread discussing this...

 

[jpiniero]

https://videocardz.com/newz/amd-ryzen-200-sku-list-with-zen4-hawk-point-silicon-leaks-out

That's a lot of models without the NPU. AI hype already dying?

 

[igor_kavinski]

Guess AI ain't for plebs...

 

[LightningZ71]

It's possible that they are just choosing to disable them to increase yields. Hawk Point will broadly compete with Kraken on most metrics, but they need Kraken to have higher ASPs. The NPU will be the differentiator.

If they were really smart, they'd have done a Hawk Point part that, instead of having an NPU, had a 16MB LLC or just an infinity cache for the iGPU. That would have done well for them in handhelds in addition to the lower end of the laptop market.

 

[jpiniero]

It's possible that they are just choosing to disable them to increase yields. Hawk Point will broadly compete with Kraken on most metrics, but they need Kraken to have higher ASPs. The NPU will be the differentiator.

If they were really smart, they'd have done a Hawk Point part that, instead of having an NPU, had a 16MB LLC or just an infinity cache for the iGPU. That would have done well for them in handhelds in addition to the lower end of the laptop market.

The purpose of the NPU is so that OEMs can claim they are selling more "AI PCs".

 

[Joe NYC]

It's possible that they are just choosing to disable them to increase yields. Hawk Point will broadly compete with Kraken on most metrics, but they need Kraken to have higher ASPs. The NPU will be the differentiator.

If they were really smart, they'd have done a Hawk Point part that, instead of having an NPU, had a 16MB LLC or just an infinity cache for the iGPU. That would have done well for them in handhelds in addition to the lower end of the laptop market.

It would need a new die and new verification to add LLC, so more trouble than the benefit AMD will get out of these in terms of revenue.

 

[StefanR5R]

The purpose of the NPU is so that OEMs can claim they are selling more "AI PCs".

If AI PC is synonymous with Copilot+ PC, then Hawk Point's NPU doesn't cut it. Maybe they indeed fuse off the NPU in Ryzen 200s like they already did in some Ryzen 8000U, or maybe they leave it alone but just don't put "AI" in the name…?

 

[igor_kavinski]

Maybe they indeed fuse off the NPU in Ryzen 200s like they already did in some Ryzen 8000U, or maybe they leave it alone but just don't put "AI" in the name…?

Or they are meant to be paired with cheap RDNA4 mobile GPUs that may offer far more TOPS than the 300 series HX NPU.

 

[StefanR5R]

Yet only NPU TOPS matter for Microsoft's Copilot+ PC sticker, not GPU TOPS, from what I understood.

 

[igor_kavinski]

Yet only NPU TOPS matter for Microsoft's Copilot+ PC sticker, not GPU TOPS, from what I understood.

Jensen said he was working with M$ to enable at least Geforce AI support in Win11. Hopefully, Lisa has something similar going on with Microsoft AI team.

 

[LightningZ71]

It would need a new die and new verification to add LLC, so more trouble than the benefit AMD will get out of these in terms of revenue.

Oh, I know it was wishful thinking. They would have had to have made that decision years ago and validated it in parallel with the NPU version.

 

[igor_kavinski]

Oh, I know it was wishful thinking. They would have had to have made that decision years ago and validated it in parallel with the NPU version.

Someone out there has an early engineering sample of that. Probably some lucky pasty white techie at a huge corporation in a basement somewhere...

 

[LightningZ71]

I doubt that it even made it to a test fab. It probably existed in the modern day equivalent of a verilog simulator and got no farther. I don't doubt that there may exist a few test samples of Strix Point in it's original configuration however.

 

[StefanR5R]

Jensen said he was working with M$ to enable at least Geforce AI support in Win11. Hopefully, Lisa has something similar going on with Microsoft AI team.

That might happen once the Copilot+ PC sticker's strategic value is diminished. E.g. when Microsoft developed a next sticker.

 

[Doug S]

Yet only NPU TOPS matter for Microsoft's Copilot+ PC sticker, not GPU TOPS, from what I understood.

Because then plenty of existing PCs would qualify as AI PCs, diminishing the ability for Microsoft to push new PC sales (i.e. new Windows license sales)

 

[jpiniero]

Because then plenty of existing PCs would qualify as AI PCs, diminishing the ability for Microsoft to push new PC sales (i.e. new Windows license sales)

If MS was concerned about that, they could limit it to 50 Series laptops only.

 

[blckgrffn]

If MS was concerned about that, they could limit it to 50 Series laptops only.

Oh you think that MS would just randomly create hardware requirements based on how new the chips are and not what they can do, huh????

 

[MadRat]

I never liked how a 'copyright protected' OS that advertises itself as fast and secure can be taken out of use simply by claiming the next OS is faster and more secure. The only way more complex systems are faster is by freezing features to handicap its use.