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

Page 323 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Timorous

Golden Member
Oct 27, 2008
1,619
2,782
136
Lets do some actual speculation.

Using this review as a baseline given it is the most recent CPU review I can find and using the 7700X as the baseline vs 8c Zen 5 then if the 40% SPECInt claim is accurate and that is also the average gaming uplift then in this suite the 8c Zen 5 would be at 121.38% with an average FPS of 279.

If the GB6 score of 3600 ST is accurate and represents an approx 23% uplift vs stock 7950X's and that applies to games then we would have the 8C Zen 5 part at 108% and 245.1 FPS.

So the range seems to be from a smidge faster than the 7800X3D (and not worth an upgrade) to quite a lot faster than the 7800X3D and for those who want the best a worthwhile upgrade.

If the $500 is true then it would probably need atleast a 25% bump in performance. That may seem like a lot and if you already purchased AM5 it may put you off upgrading at launch but if you look at platform costs there are now cheaper AM5 motherboards and DDR5 is cheaper so someone upgrading from AM4 may not spend a lot or any more on a 9800X rig than someone spent buying a 7700X rig at launch. The difference is AMD are taking a larger portion of that pie.

If the 9800X hits the upper end of this performance range then I can easily see $499 as the asking price. It probably would not impact overall platform cost all that much vs a new build when the 7700X launched so why not take more of the pie. If it is towards the lower end of this range then the price will probably need to be closer to $450 and going 7800X3D for gamers while they are still around would be the no brainer move if they don't want to wait for 9800X3D.
 
  • Like
Reactions: Tlh97 and Joe NYC

cortexa99

Senior member
Jul 2, 2018
319
505
136
Internal information that StrixPoint is 1.3x MT perf of HawkPoint/Phoenix.

I'm not sure but just calculate by yourself. 8700G MT scores ~18000pts in R23 and 7940HS is ~17000pts in some models.
 

gdansk

Platinum Member
Feb 8, 2011
2,123
2,629
136
Why are all of you assuming those GB numbers are correct? What is the source, RGT?
I can assure you they are made up as I typed them myself. But really I don't expect much higher. GB6 puts 7950X at 2939 average single core score, so this represents a 23% improvement.

Next time I'll add a speculation tag but I thought paralleling this junk estimate from the Apple SoC thread was clear.

Apple A18 Pro Geekbench score leak:
View attachment 95865
A solid 20% jump in Single-core.
 
Last edited:

gdansk

Platinum Member
Feb 8, 2011
2,123
2,629
136
Great. Just triggered two pages of confusion and incorrect speculation in the wrong direction.
Are you sure it's the wrong direction? I'm pretty sure you speculated it could be even lower at times.
And I should have let them run with it because at least it is a realistic baseless number instead of the exceptional 32% and now 40%.
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,787
256
126
Member callouts are not permitted
Are you sure it's the wrong direction? I'm pretty sure you speculated it could be even lower at times.
And I should have let them run with it because at least it is a realistic baseless number instead of the exceptional 32% and now 40%.
I don’t know whether it’s a likely prediction or not. But it would have been good to know that it was just your personal guesstimate, and not an actual measurement from a Zen5 CPU which was how it was intepreted by people in this thread.

Anyway, what makes you think GB6 score you speculated at 3628 is close to the correct one? It means approximately 20% perf increase. And since others mentioned it correlates well with SIR 2017 ST, it would mean ~20% in that benchmark too. So indirectly it also means that the 40+ % perf increase in SIR 2017 ST claimed by e.g.redacted is incorrect.
 
Last edited by a moderator:

Fjodor2001

Diamond Member
Feb 6, 2010
3,787
256
126
12 core vs 8 core CPU.
More precisely 4xZen5 + 8xZen5C vs 8xZen4.

If it would have been 12xZen5, it would mean a regression core-to-core if MT perf increase over 8xZen4 was only 30%.

Since Strix Point is a mix of Zen5 & Zen5C it’s hard to draw any conclusions regarding Zen5 perf when comparing MT perf of Strix Point vs pure Zen4 CPUs.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,821
3,643
136
Internal information that StrixPoint is 1.3x MT perf of HawkPoint/Phoenix.

I'm not sure but just calculate by yourself. 8700G MT scores ~18000pts in R23 and 7940HS is ~17000pts in some models.
Assuming that they are working within the same power envelope, and speculating the Zen5c cores are 60% perf of the Zen5 cores, a 1.3x improvement in MT score in CB r23 can be explained by a 20% increase in per core performance of Zen5 over Zen4 in that specific workload.
 

Timmah!

Golden Member
Jul 24, 2010
1,429
648
136
Tell me the CAD performance, thats what i care about. Viewport performance on AutoCAD is everything but fluent on 7950x - on drawings with lot of blocks and whatnot. In that regard not much changed compared to my previous Intel Skylake-X cpu.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,821
3,643
136
Tell me the CAD performance, thats what i care about. Viewport performance on AutoCAD is everything but fluent on 7950x - on drawings with lot of blocks and whatnot. In that regard not much changed compared to my previous Intel Skylake-X cpu.
Viewport performance is limited by ST perf.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,362
2,854
106
More precisely 4xZen5 + 8xZen5C vs 8xZen4.

If it would have been 12xZen5, it would mean a regression core-to-core if MT perf increase over 8xZen4 was only 30%.

Since Strix Point is a mix of Zen5 & Zen5C it’s hard to draw any conclusions regarding Zen5 perf when comparing MT perf of Strix Point vs pure Zen4 CPUs.
Zen5 vs Zen5C should be like Zen4 vs Zen4C.
IPC the same, but lower clocks.
Only 30% increase should be most likely due to a limited TDP.
Higher IPC + 50% more cores shouldn't bring only 30% more nT performance.
 
  • Like
Reactions: Tlh97