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

Page 618 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Sort by date Sort by votes
poke01

poke01

Platinum Member
Mar 8, 2022
2,157
2,748
106
  • Like
Reactions: Mopetar and igor_kavinski
Upvote 0 Downvote
T

tsamolotoff

Member
May 19, 2019
183
315
136
igor_kavinski said:
Yes. Yes. Give up. Give into dismay. That is the way. That is how you get closer to appeasing Darth Lisa with your fealty
Click to expand...
Apathy is death, yada yada.

It's not dismay, I've already did full power run and posted the result here, but I think it was removed by either mods or some automod (as it contained a link to blender benchmark database). Thanks to x3d limits, CPU doesn't go above 1.1v in multicore loads (it is usually well below that), so we have to keep faith in our cacheless brethren
 
  • Haha
Reactions: igor_kavinski
Upvote 0 Downvote
gaav87

gaav87

Member
Apr 27, 2024
122
163
76
tsamolotoff said:
I can say without even running it that my CPU would be below that, I think total score for full power run was around 580 (if you sum all these values). It is actually interesting whether Z5 scales beyond these 160W, as it seems Z4 starts to face severe diminishing returns at this cutoff
Click to expand...
Slug results:
7950x ~617,38 / 230W no-PBO
This ~678,1 / 160W ~10% increase
~715 minimum at 230w seems resonable if it has same clock eff. as 7950x
 
  • Like
Reactions: lightmanek
Upvote 0 Downvote
Hail The Brain Slug

Hail The Brain Slug

Diamond Member
Oct 10, 2005
3,507
2,459
136
Det0x said:
But sadly its still alittle unclear to me if the 7950X3D are benefiting from the v-cache on CCD0 in blender, or if its all down to voltage tuning for that SKU, since it seems to outperform the 7950X in the powerlimited scenario we're looking at in this thread.
Click to expand...
My 7950X clocks 2.5-2.7GHz effective clocks at 60W PPT, ~30-31W core power.
My 7950X3D clocks 3.0-3.2GHz effective clocks at 60W PPT, ~28-29W core power.

The V$ might help some but it's mostly the V/F curve. If I undid the memory tuning on my 7950X3D and went straight EXPO I'd have a couple more watts to increase clocks a little more.
 
  • Like
Reactions: igor_kavinski, lightmanek, Tlh97 and 3 others
Upvote 0 Downvote
Josh128

Josh128

Senior member
Oct 14, 2022
363
510
96
Zen 5 IHS shows 2023 date. Disqus commenter Senpai noted the visible code seems to indicate 15th week of 2023. Thats over a year ago. Strange. Maybe because its got an old IOD?

videocardz.com

AMD starts shipping Ryzen 9000 review samples - VideoCardz.com

AMD now providing samples of Zen5 CPUs to reviewers The testing has begun. AMD has not yet confirmed the release date for its Ryzen 9000 series, featuring Zen5 cores. Meanwhile, AMD is working behind the scenes to provide samples for reviewers, whose feedback will be crucial for enthusiasts...
videocardz.com videocardz.com
 
Upvote 0 Downvote
Josh128

Josh128

Senior member
Oct 14, 2022
363
510
96
Wild speculation here. Turin is 3nm 16 core Zen 5c as far as we know. Based on MLIDs rather legit looking slide that called for 10-15% IPC for Zen 5, 4&3 nm variants, it also called out a new 16 core CCX. Guessing thats what Turin is.

What if that 3nm 16core CCX CCD was designed with X3D vias on it? Wouldnt it be something if at some point they surprise with a 16 core unified L3 X3D 3nm chip? Depending on clocks, seems like it could be a potent SKU.
 
  • Love
Reactions: igor_kavinski
Upvote 0 Downvote
S

StefanR5R

Elite Member
Dec 10, 2016
5,920
8,839
136
igor_kavinski said:
They held it back!!! :eek:
Click to expand...
As AMD said to TPU, they are a software shop now. Perhaps they want to debut Zen 5 with a really stable AGESA at launch, for once.

Josh128 said:
Wild speculation here. Turin is 3nm 16 core Zen 5c as far as we know. Based on MLIDs rather legit looking slide that called for 10-15% IPC for Zen 5, 4&3 nm variants, it also called out a new 16 core CCX. Guessing thats what Turin is.

What if that 3nm 16core CCX CCD was designed with X3D vias on it? Wouldnt it be something if at some point they surprise with a 16 core unified L3 X3D 3nm chip? Depending on clocks, seems like it could be a potent SKU.
Click to expand...
3nm 16c CCX is for Turin-dense only (are the rumors saying). Turin proper is 4nm 8c CCX like Granite Ridge (but allegedly at a higher stepping). V-cache is for the classic CCDs only, not for the dense ones, just like in the current 9004 generation (Genoa/ Genoa-X, vs. cache-reduced Bergamo and Siena).
 
  • Like
Reactions: Tlh97 and SarahKerrigan
Upvote 0 Downvote
Josh128

Josh128

Senior member
Oct 14, 2022
363
510
96
StefanR5R said:
As AMD said to TPU, they are a software shop now. Perhaps they want to debut Zen 5 with a really stable AGESA at launch, for once.


3nm 16c CCX is for Turin-dense only. Turin proper is 4nm 8c CCX like Granite Ridge (but allegedly at a higher stepping). V-cache is for the classic CCDs only, not for the dense ones, just like in the current 9004 generation (Genoa/ Genoa-X, vs. cache-reduced Bergamo and Siena).
Click to expand...

Sure, thats how it was done last gen, but whos to say it wasnt designed with this option in mind for the future? 16 core Zen 4 CCD was already possible on 5/4nm. 3nm should be quite a bit more dense. Whats the extra density used for? Just smaller silicon and reduced power only?
 
Upvote 0 Downvote
S

StefanR5R

Elite Member
Dec 10, 2016
5,920
8,839
136
Josh128 said:
Whats the extra density used for?
Click to expand...
I don't know. But to me, it seems reasonable to assume that there will be only a single 16c CCX physical design in the Zen 5 generation, and that this one will be for "cloud native" server products primarily. For client (desktop in particular), a physical design needs to allow for high peak clocks. In contrast, a design which is supposed to be targeted to hyperscalers first and foremost will trade peak clocks for density.

AMD is supplying the desktop with server-derived designs and mobile-derived designs. Ultimately, that's what the other desktop CPU makers do too. Don't hold your breath for a high core count desktop exclusive design.
 
  • Like
Reactions: Tlh97 and Thunder 57
Upvote 0 Downvote
M

MarkPost

Senior member
Mar 1, 2017
313
591
136
  • Like
Reactions: Tlh97, Gideon, Schmide and 3 others
Upvote 0 Downvote
S

StefanR5R

Elite Member
Dec 10, 2016
5,920
8,839
136
@Josh128, PS: Of course I too am curious how their 16c CCX (and the alleged Zen 6 32c CCX) is going to look like. But one thing is for certain: The larger a core complex topology is, the higher the lastlevel cache latencies, and with it the core-to-core latencies, are going to be. The shrink to 3nm may help in this regard though.
 
Upvote 0 Downvote
Hail The Brain Slug

Hail The Brain Slug

Diamond Member
Oct 10, 2005
3,507
2,459
136
  • Like
Reactions: Tlh97, Gideon, Kryohi and 3 others
Upvote 0 Downvote
Hail The Brain Slug

Hail The Brain Slug

Diamond Member
Oct 10, 2005
3,507
2,459
136
Nothingness said:
It's not the copyright date but the mention "BY 2315PGY" that's strange. That would indicate 15th week of 2023.
Click to expand...
Very strange there would even exist a retail etched sample from so long ago. It doesn't seem like it's necessarily one of the review samples, that article is kind of all over the place saying bitwit kyle "found it on a plane". The photos of the actual review samples have the details blurred on the IHS.

I checked, Zen 4 reviews had ~2226 date samples, so only ~13 week old samples. But this sample appears to be 65 weeks old. Over a year older than the review samples from last time.
 
  • Like
Reactions: Tlh97, lightmanek and Nothingness
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom