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

Page 708 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Diamond Member
Apr 15, 2012
6,187
10,696
136
AT posted surprisingly high gain on gcc subtest of SPEC, tough compared to 7700, but it's 17 %

Edit:
They have also significantly updated details about the test environment for SPEC, newest CLANG, enabling AVX512 and specifying the WSL version used. Nice:)

AT SPECint showing 13% improvement overall. The 9700x has a slight clock speed advantage according to boost specs, but without knowing actual running clock speeds, we can't calculate IPC. It is 9% improvement if we assume spec boost speeds, but With Zen 4/5, that's not a safe assumption (could be higher than spec). SPECfp showed significantly higher improvement at 26% overall. For consumer purposes, too many resources were used to improve FP versus INT. For some server customers, maybe AI (?), this will be really good.

136699.png
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136

branch_suggestion

Senior member
Aug 4, 2023
392
875
96
Phoronix review shows the thing is actually very good, just not in the ways we hoped.
Geomean score is similar to a 13900K with half the power consumption.
As it turns out, FP gains are significantly higher than INT, something I guessed ages ago just from reading the uArch changes.
Oh, and the biggest takeaway is the most significant gains are in traditionally Intel dominated workloads.
And Linux results in general just look way better than Windows, why exactly is a question for AMD software engineers and whichever intern Microsoft sends as a sacrifice.
 

leoneazzurro

Golden Member
Jul 26, 2016
1,052
1,716
136
AT SPECint showing 13% improvement overall. The 9700x has a slight clock speed advantage according to boost specs, but without knowing actual running clock speeds, we can't calculate IPC. It is 9% improvement if we assume spec boost speeds, but With Zen 4/5, that's not a safe assumption (could be higher than spec). SPECfp showed significantly higher improvement at 26% overall. For consumer purposes, too many resources were used to improve FP versus INT. For some server customers, maybe AI (?), this will be really good.

136699.png
Hwupgrade review found a substantial clock difference between the 7700X-7600X and 9700X-9600X, because the TDP holds back the performance for these SKUs. In Cinebench they foung 500MHz all-core clock difference between the 7700X and 9700X, with the 7700X consuming 40W more, and the 9700X was slightly faster.
 

CouncilorIrissa

Senior member
Jul 28, 2023
542
2,124
96
Reality: Zen 10% at greater efficiency.
Harsh Reality: Lowered TDP requires PBO on to see performance improvements in many cases.

The browser benchmarks from Phoronix are so interesting. I don't know what sets these tests a part so much. From what I understand, browsing can be very branchy and hard to predict, but I don't know if that holds true for these scripted benchmarks. Maybe the improved branch predictor and dual decoder front end shines here?

View attachment 104651
View attachment 104652
BPU is great if you look at the tests done by Huang.
It's something else that holds everything back.
 

Hitman928

Diamond Member
Apr 15, 2012
6,187
10,696
136
Hwupgrade review found a substantial clock difference between the 7700X-7600X and 9700X-9600X, because the TDP holds back the performance for these SKUs. In Cinebench they foung 500MHz all-core clock difference between the 7700X and 9700X, with the 7700X consuming 40W more, and the 9700X was slightly faster.

That's all core load. I'm talking single core clocks.
 
  • Like
Reactions: Tlh97 and yuri69

branch_suggestion

Senior member
Aug 4, 2023
392
875
96
The worst part is that we’ll be stuck with this total dud of a core for years and Zen6 will only bring iterative improvements over it. At least improvements to the uncore will be something to look forward to.
Z6 is by the Z4/Z2 team, they know how to iterate on an existing core, just hope they get a nice bump in area to play with.
If they can reintroduce a lot of the Z4 refinements and optimisations than Z5 is missing that would be fantastic.
 
  • Like
Reactions: exquisitechar

gdansk

Platinum Member
Feb 8, 2011
2,967
4,498
136
Looking like a long reign for the 7800X3D gaming champ.

Other tests show web browser performance improvement. I guess they did figure out a way to target improved interpreter and JIT performance.
 
  • Like
Reactions: lightmanek

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,441
5,434
136
From AT's review - there are some workloads (e.g. AVX512) where the new architecture truly shines...
1723039179881.png

My take:
At stock TDP, comparing 9700X to vanilla 7700 gives a more accurate generational Zen 4 v 5 comparison

The good:
Power efficiency is WAY better than my 7700X stock - there's a reason why I ran it in ECO mode
Some workloads (like above) show huge leaps in performance vs Zen 4

The bad:
SEP prices still a tad high given the lower absolute increases in performance IMO. We all know that a few months after launch prices will trend down so I expect it will be on sale before Xmas

The ugly:
I think we've seen for multiple generations now (both CPU and GPU) that it's generally not worth it to upgrade every gen. The low-hanging fruit for improvement has been plucked already and physical limits are making it harder to get the big gains we all want. Which means waiting at least 2 gens for a meaningful upgrade.
 

gdansk

Platinum Member
Feb 8, 2011
2,967
4,498
136
What I'm really wondering is if the X3D cache uplift will be as much as the previous generation, it might be somewhat less if its the exact same cache.
I think the TSVs are in different locations so it is probably not the exact same cache die. Whether it ends up with more capacity or not is an interesting question.
 
  • Like
Reactions: Joe NYC