Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[poke01]

I laugh at M4 Max! As you’ve seen already, this is my M4 in Chrome 130. 🤓

View attachment 111494

Nice, I saw the M4 reach over 500 in one of the YouTube videos. I’ll post in the web thread when I find that video again.

 

[stayfrosty]

AMD Announces Versal Premium Series Gen 2 Enabling New Levels of System Acceleration to Meet the Demands of Data-Intensive Workloads

Extends Versal Gen 2 portfolio with industry's first FPGA devices featuring CXL 3.1 and PCIe Gen6, with LPDDR5X support, for fast connectivity,…...

ir.amd.com

With support for CXL 3.1 and PCIe Gen6, the industry’s fastest host interfaces,2 Versal Premium Gen 2 devices enable industry-leading, high-bandwidth host CPU-to-accelerator connectivity.

AMD Versal Premium Series Gen 2 development tools are expected to be available in Q2 2025, followed by the availability of silicon samples by early 2026. Production shipments are expected to begin in the second half of 2026.

There's no way AMD isn't coinciding a PCIe Gen6 ready DC Platform with this. Venice / Zen 6 -> H2 2026 confirmed? 😆

 

[Hitman928]

AMD Announces Versal Premium Series Gen 2 Enabling New Levels of System Acceleration to Meet the Demands of Data-Intensive Workloads

Extends Versal Gen 2 portfolio with industry's first FPGA devices featuring CXL 3.1 and PCIe Gen6, with LPDDR5X support, for fast connectivity,…...

ir.amd.com

There's no way AMD isn't coinciding a PCIe Gen6 ready DC Platform with this. Venice / Zen 6 -> H2 2026 confirmed? 😆

There was this:

 

[SteinFG]

There's no reason for 6.0 on client devices so soon, mainstream GPUs are getting 5.0 only in 2025. It might take 4 years before we see 6.0

AM6 in 2028 is a better bet for this

 

[Eug]

Nice, I saw the M4 reach over 500 in one of the YouTube videos. I’ll post in the web thread when I find that video again.

Psychological barrier broken. I tried it again and this time I got 492. No 500 for me. However, it’s not a totally clean system. I have a few extra services running in the background.

For context, this M4 is my work machine, and I mainly just use business office type applications, so even something with just half the CPU performance would be totally fine. It’s interesting to note, however, that a lower performance machine is not actually available for purchase new.

 

[poke01]

https://videocardz.com/newz/s-t-a-l-k-e-r-2-heart-of-chornobyl-pc-requirements-revealed-dlss3-and-fsr3-1-at-launch

This is the perfect game to test the 9800X3D to showcase its capabilities.

 

[RnR_au]

Strix Halo makes most sense for one group of people and that’s LLM users.

idk.. I used to think the same. Then I picked up old mining cards with twice the membw of what Strix Halo is supposed to have. 40GB of vram for $AU400 landed. So now I have a small gpu cluster in the works which I'll VPN into.

If an LLM user really wants to run models on a laptop they'll max out an Apple unit.

In the end membw is king for inference, and Strix Halo simply don't have enough to be real interesting. Additionally, I don't expect any of the AMD partners to ship 128GB sku's.

 

[dr1337]

A 40 CU Strix mini PC won’t be cost effective as a steam Console. Better off building an actual SFF PC. More modular.

Strix Halo makes most sense for one group of people and that’s LLM users. Gamers are better served by mobile RDNA4/RTX 50 laptops/SFF PCs.

Meh, I bet it won't be that expensive to the OEMs. Especially in a world where we have all these different companies making x86 gaming handhelds. And it should be a good bit cheaper than current gaming laptops having to deal with a dGPU.

And yeah the market bleeds green on gaming laptops, but if AMD sells strix halo powered SKUs at the same cost as 4060m devices while being 20% faster and more efficient, that could be attractive to consumers.

I do agree that LLM customers will like 64 and 128gb SKUs for sure, especially if the OEMs price them lower than apple.

 

[jpiniero]

Meh, I bet it won't be that expensive to the OEMs. Especially in a world where we have all these different companies making x86 gaming handhelds.

Look at how expensive just Strix Point is.

 

[ToTTenTranz]

Strix Halo makes most sense for one group of people and that’s LLM users. Gamers are better served by mobile RDNA4/RTX 50 laptops/SFF PCs.

Systems with dGPUs will hardly be able to compete with Strix Halo in performance per watt below 75W.

This means smaller systems that can't have larger heatsinks will benefit from Halo's integration.

 

[FlameTail]

Systems with dGPUs will hardly be able to compete with Strix Halo in performance per watt below 75W.

This means smaller systems that can't have larger heatsinks will benefit from Halo's integration.

If the dGPU is wide enough (eg: RTX 5080) configured to run at 75W, I think it will be more efficient than Strix Halo in active heavy use.

Where Strix Halo would have a definitive advantage is in idle power and what that entails for battery life.

 

[Thibsie]

5080@75W ?
Is that even possible ?

 

[Ghostsonplanets]

5080@75W ?
Is that even possible ?

Yes. Nvidia allows OEMs to set their dGPU TGP targets. It's not uncommon to buy a thin and light gaming laptop with xx80 card restricted to 70/80/90W TGP.

 

[Tuna-Fish]

5080@75W ?
Is that even possible ?

Just to note that mobile 4080 is AD104, not AD103. It's configurable down to 60W, at which point it runs with a baseclock of <800MHz, and is close to the performance of desktop 4060.


5080 will likely be similar. Most laptops will likely configure it closer to it's optimal point.

 

[Thibsie]

Ah, was stuck to discreet 5080 somehow.
Got it, thanks.

 

[inquiss]

The ship has sailed on Zen 5 already. I think the demand will be down for 9800x3d after a week. They really should set the price at $399. People building new systems or buying new computers have largely left the PC market.

So how are we feeling about this now? How do you feel about the product post review?

 

[StefanR5R]

He didn't say after which week.

 

[sl0519]

So how are we feeling about this now? How do you feel about the product post review?

This post aged like milk~

 

[inquiss]

This post aged like milk~

Yup, they'd have been sold out I expect with a much higher price. The best of the best gets to charge the premium

 

[sl0519]

Yup, they'd have been sold out I expect with a much higher price. The best of the best gets to charge the premium

Imagine AMD dominated Nvidia 3 generations in succession (that's all it takes for AMD to take over intel for 3 generations of X3D) and Nvidia has to lower its price to compete. I'm daydreaming lol.

 

[Hans Gruber]

So how are we feeling about this now? How do you feel about the product post review?

Pretty good. I still think there is more performance to be gained and better efficiency through bios updates. AMD is rapidly fixing their Zen 5 issues with the 9800x3D.

 

[inquiss]

Pretty good. I still think there is more performance to be gained and better efficiency through bios updates. AMD is rapidly fixing their Zen 5 issues with the 9800x

what about the main part of the message? I'm only joshing here by the way. Meant to be in good humour.

 

[Hans de Vries]

very speculative question:

in what range is the 'gift' from Intel to Userbenchmarks? 6 digits?

I don't know, but I do not think so...

Userbenchmark is made and maintained by just a single 46 year old deaf free-lance web designer. He's looking for work to support his familie.

 

[Josh128]

I don't know, but I do not think so...

Userbenchmark is made and maintained by just a single 46 year old deaf free-lance web designer. He's looking for work to support his familie.

Is he a software engineer as well or is there someone else that designs the benchmark software? He needs to be careful as his bias mixed with running a benchmarking software and site is not a good thing if he's worried about supporting his family. If he's not getting bankrolled by Intel for that bias, he may have some mental issues.

 

[jpiniero]

Look at how expensive just Strix Point is.

Case in point - Golden Pig said like a week or two ago predicting that Blackwell gaming laptops are going to be basically Raptor Lake based... and maybe Phoenix/Hawk or perhaps even older. I see no reason to think he's wrong on that with the CPU price being an issue.

AMD is even rebranding Hawk into Ryzen 200 series:

https://videocardz.com/newz/amd-preparing-ryzen-7-255-260hs-a-re-refresh-of-hawk-point-8745h-8845h-skus

AI Hype has given AMD an opening but it's tough to say how big of a market this is. I wouldn't expect to see much of Strix Halo being marketed for gaming. For AMD+AMD I would expect Hawk+RDNA4 if AMD can manage to get some deals this time.