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

Page 946 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Mar 11, 2004
23,441
5,841
146
The big, unanswered question for me with APUs equipped with nontrivial NPUs is about memory bandwidth. Most APUs out there are living in a cost controlled environment that heavily constrains memory bandwidth. Strix Point has a decently beefy iGPU and is saddled with lower spec "dual" channel DDR5. Similar goes for Arrow Lake H. Even Strix Halo with it's 256 but memory interface is still getting saturated with a 40cu iGPU. Where's the extra memory bandwidth going to come from to keep the NPU properly fed? When the NPU gets used for anything but the most trivial tasks, does the frame rate collapse because it's fighting with both the iGPU and the CPU cores for what scraps of bandwidth are available?

This is one area where I see Apple having a slight advantage as they tend to provide a bit more memory bandwidth compared to the capabilities of their processor specs. Performance should fall off a bit more gracefully there as all three subsystems get loaded.

Exactly, using the NPU is gonna take resources from other parts. But hey, just sign up for Xbox ALive so you can keep your chatbot smart and your gameplay uninterrupted. I wonder what streamer they'll sign as their first AI Xbox voice so you can hear them scream slurs at you while their character teabags you.
 
Reactions: Thibsie

fastandfurious6

Senior member
Jun 1, 2024
440
588
96
0. across society and industries "creation" is plateauing, most new movies games music etc is hot trash

1. games of recent times are increasingly terrible, even AAA. Art and meaning are getting lost

2. using AI for ingame content generation (stages, voice, objects, w/e) exacerbates this, games will feel even more fake

3. using AI for "performance increase" is a pipe dream pushed by those not understanding how performance works

did I miss anything
 

DrMrLordX

Lifer
Apr 27, 2000
22,491
12,364
136
Press X to pay respect!

Can't I press F instead?

Or better yet, they'll let you change your voice, so all the pedophiles and white supremacists can groom kids on Xbox Live!

They're all on Discord. Or Roblox.

At this point it seems that you're not all that impressed with NPUs, and to be honest, I still don't see a purpose for them either. Maybe someone will make good use of them eventually. For now, iGPUs and dGPUs have so much more inference power that NPUs seem like a waste of silicon.
 

Hail The Brain Slug

Diamond Member
Oct 10, 2005
3,709
2,963
136

Thunder 57

Diamond Member
Aug 19, 2007
3,418
5,655
136
Jul 27, 2020
23,535
16,533
146
True, but in that case they want you to buy an Epyc, not a "cheap" desktop part.
Strix Halo is now a mini desktop part and NOT cheap

I think dual CCD V-cache will come true someday. Someone at AMD is plotting as we speak on how to make it happen, just like someone did for the ultimate APU aka Strix Halo.
 

Thunder 57

Diamond Member
Aug 19, 2007
3,418
5,655
136
Strix Halo is now a mini desktop part and NOT cheap

I think dual CCD V-cache will come true someday. Someone at AMD is plotting as we speak on how to make it happen, just like someone did for the ultimate APU aka Strix Halo.

I think so too. But right now they have trouble enough producing 9800X3D's. Maybe once production isn't an issue they will.
 

LightningZ71

Platinum Member
Mar 10, 2017
2,077
2,525
136
True, but in that case they want you to buy an Epyc, not a "cheap" desktop part.
The cheapest X series Epyc Zen4 is listed at about $5000, and that's before a single cent of platform costs, for a 16 core processor. A 9950X3d2 would be profitable at $1000 and have nearly the same performance in many loads (but certainly not all and much more in others with higher clocks). Sell it under the Epyc 4xxx line as the 4685PX if you have to.
 
Reactions: Win2012R2

inquiss

Senior member
Oct 13, 2010
352
527
136
I think so too. But right now they have trouble enough producing 9800X3D's. Maybe once production isn't an issue they will.
Exactly this. Stacking is the limiting factor for production. Though I expect it to remain that way for a long time. HBM would be useful for graphics, but it's going to be capacity limited for forever. Same for stacking. A dual CCD stacked client part isn't anywhere near the top of the tree for importance, I expect it to similarly be unavailable for a long time
 

fastandfurious6

Senior member
Jun 1, 2024
440
588
96
there's bound to be some slowdown in fab demand as LLMs become less resource-intensive and shops already have billions of GPUs in stock

also computing is already at a point where upgrades from now on will become more and more meaningless

there's bound to be increasing tons of unsold stock

also on mobile phones it's now already at least 4-5 years where upgrades have become meaningless, for example Galaxy S26 offers very little over Galaxy S21 or whatever

https://www.gsmarena.com/compare.php3?idPhone1=13610&idPhone2=10626#diff- literally the only main diff is 3nm vs 5nm chip over 4 years and it's a completely meaningless upgrade
 

deasd

Senior member
Dec 31, 2013
592
1,011
136
Any confirmation that Zen5 cores in StrixHalo has full AVX512 width FPU or not? I can only remember Zen5 in StrixHalo looks similar to GraniteRidge.
 
Mar 11, 2004
23,441
5,841
146
Strix Halo is now a mini desktop part and NOT cheap

I think dual CCD V-cache will come true someday. Someone at AMD is plotting as we speak on how to make it happen, just like someone did for the ultimate APU aka Strix Halo.

I feel like stacking it on the memory controller means they'd only need one still, and could benefit multiple dies (be they CPU or GPU).

I wondered if that wasn't even the idea behind the c cores to begin with, they go with limited L3 because they'd be stacking it. Beef up L1 and L2 instead.

The cheapest X series Epyc Zen4 is listed at about $5000, and that's before a single cent of platform costs, for a 16 core processor. A 9950X3d2 would be profitable at $1000 and have nearly the same performance in many loads (but certainly not all and much more in others with higher clocks). Sell it under the Epyc 4xxx line as the 4685PX if you have to.

I'm still surprised that AMD didn't just push the multi dice chips to Threadripper. Especially this year with how the 370 chip could cover most consumer wants. I'm also surprised they haven't done something like make the Zen 5c dice be the X3D chip (sacrifice some of the density towards the TSV to enable the stacked cache), which would make them more appealing as you'd gain the cache for games but gain cores as well. It'd let them simplify the stack, or maybe even add a 3rd dice, so they do single dice at $599 where it'd offer what the 9900X3D does, a dual dice at $799 which would offer more than the 9950X3D, and then a third top of the line one at $999.

I really feel like AMD should have gone APU with embedded memory for consumer (the 370 AI chip really would've been good enough, especially with fast RAM), merge AM5 and Threadripper as a single platform go quad channel. They'd be able to offer the core counts of Threadripper outside the highest (which is what EPYC is for), plus it could've offered the ability to do Strix Halo like chips with even more RAM.

there's bound to be some slowdown in fab demand as LLMs become less resource-intensive and shops already have billions of GPUs in stock

also computing is already at a point where upgrades from now on will become more and more meaningless

there's bound to be increasing tons of unsold stock

also on mobile phones it's now already at least 4-5 years where upgrades have become meaningless, for example Galaxy S26 offers very little over Galaxy S21 or whatever

https://www.gsmarena.com/compare.php3?idPhone1=13610&idPhone2=10626#diff- literally the only main diff is 3nm vs 5nm chip over 4 years and it's a completely meaningless upgrade

Until they push larger AI chips in every internet of things devices and other claiming they need it to run all those "less resource intensive LLMs" on device, and suddenly every single thing (including toasters and probably the kitchen sink) need to have SoCs in them.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |