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

Page 972 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ToTTenTranz

Senior member
Feb 4, 2021
686
1,147
136
A 9070XT is a large die with 640 GB/s of membw available to it, combined with 64MB of IC. You're not getting anywhere close to it with an APU.

Probably not on TSMC N3, but on N2 I could see this happening. Especially as N2 is finally bringing some promising advances on memory cell area (i.e. higher cache density).

Besides if there's a PS6 coming out, it's safe to assume it'll have a more powerful iGPU than the 9070XT.
 

LightningZ71

Platinum Member
Mar 10, 2017
2,508
3,190
136
There was a bit of noise about a Rembrandt based desktop APU at one point. With it still living in OEM space still, there's always a chance.
 

eek2121

Diamond Member
Aug 2, 2005
3,408
5,046
136
I wish! 🤣

Maybe 9950X3D guaranteed 6 GHz edition :p

They have seen that the market is mad for their CPUs so they are like, hey, let's raise the bar why don't we?
A new stepping has seemingly been confirmed, so that isn't out of the realm of possibility. While we haven't had anything awesome recently, new steppings in the distant past have brought some pretty nice performance and/or overclocking improvements.
 

StefanR5R

Elite Member
Dec 10, 2016
6,670
10,546
136
AMD Granite Ridge CPU New Stepping: B40F41C1?
A new stepping has seemingly been confirmed,
Or more precisely, a new CPU ID which is the same as present Granite Ridge's CPU ID but incremented at the least significant digit, a.k.a. the "stepping" part of the CPU ID. How about a Granite Ridge with an improved cIOD? Improved WRT GMI and IMC? (E.g.: GMI-wide support, CUDIMM support; the former making the latter useful.)


Or in the worst case, with nothing but an NPU added, for MS sticker eligibility.
 

Jan Olšan

Senior member
Jan 12, 2017
572
1,129
136
Not really, that in itself would need designing a new die, which if you know AMD's CPU history of last 15 years, they prefer to really only do as a last resort.
 

StefanR5R

Elite Member
Dec 10, 2016
6,670
10,546
136
They'd do it if paid for, obviously. But (continuing to state the obvious) putting Strix Point/ Krackan/ the rumored Gorgon Point into AM5 would be the much shorter = much cheaper for AMD's customers but not equivalent path to delivering an NPU to the AM5 platform. The 'model' bits of the new CPUID are Granite Ridge's though, so let's see what's been cooking for GNR.
 

Bigos

Senior member
Jun 2, 2019
204
519
136
It already does, you're not decode-limited due to opcache anyway.
Do you have a benchmark or just "trust me bro"?. And "95% is not using opcach" is hardly an answer. I understand many workloads with small loop sizes but many != All. Saying "dual decode for single thrard is bad cause trust me bro" is inane. Just say "I have nothing to say about it" and be done with it. Or do you have data to prove it?
 

poke01

Diamond Member
Mar 8, 2022
4,196
5,542
106

This is why huge APUs are the future, imo. Perf on battery is appalling on seperate dGPU setups. 1hr to render a Premiere project while unplugged, and close to 5 minutes plugged in on the Razer. Could this be a razor issue? maybe
 
Jul 27, 2020
27,954
19,101
146

9600 MT/s support for Ryzen so most definitely CUDIMM support is incoming!