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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

and the FADD latency is higher.

They're both 2c effective.

 

[MS_AT]

They're both 2c effective.

If we are to believe measurements done by Alex Yee and David Huang only Granite Ridge is able to do 2c under specific circumstances http://www.numberworld.org/blogs/2024_8_7_zen5_avx512_teardown/#fadd_latency

 

[adroc_thurston]

If we are to believe measurements done by Alex Yee

I know what mysticial said.
Strix is also 2c effective and has the same scheduler hazard.

 

[Tup3x]

Yes. Why wouldnt it be?

Yes with cutdown AVX-512 to save die space

Zen 5 in Strix is far more different than RPC compared to GLC if anything.

It's not only the lack of 512-bit wide datapath, the vector PRF for 128- and 256-bit operations is smaller and the FADD latency is higher.

Yep. Cache or even manufacturing process means little. Zen 5 in Strix is indeed quite a bit different than on desktop (let alone monolithic die) yet it's indeed Zen 5. Considering how little has changed Intel side, it's fairly accurate to say that they are "pretty much the same thing" even though they aren't exactly the same thing.

 

[MS_AT]

I know what mysticial said.
Strix is also 2c effective and has the same scheduler hazard.

I am getting lost in the hazards, do you mean the scheduler hazard that adds 1 cycle latency to ops that are supposed to be 1 cycle only or is there some otherwise undocumented FADD specific scheduler hazard that prevents 2 cycle FADDs to be used every time? As according to all official info on the former, it shouldn't apply to FADD as FADD is anyway is at least 2 cycle op.

 

[inf64]

Another YT review, very interesting results with tuned 9700X with windows 11 updates and Core Isolation disabled (vs tuned 7800X3D)|

 

[Thibsie]

Zen 5 in Strix is far more different than RPC compared to GLC if anything.

It's not only the lack of 512-bit wide datapath, the vector PRF for 128- and 256-bit operations is smaller and the FADD latency is higher.

Was just trying to explain what I think Tup3x had in mind.

 

[LightningZ71]

Has anyone seen how Phoenix/Hawkpoint did after the windows updates and how they compare to Strix Point in ST and MT afterwards? I've been trawling the interwebz and haven't seen anything.

 

[static shock]

The IPC uplift was 50% , not 16% as it was launched, when i die they will launch the performance original BIOS. Ban me Lisa Su. Please.

 

[positivedoppler]

Another YT review, very interesting results with tuned 9700X with windows 11 updates and Core Isolation disabled (vs tuned 7800X3D)|

View attachment 107232
View attachment 107233


View attachment 107234

AMD did say they will fix the core latency issue so hopefully more performance yet to be unlocked

 

[GTracing]

AMD did say they will fix the core latency issue so hopefully more performance yet to be unlocked

Isn't the latency issue just for CPUs with multiple CCXs? The 9700X is only has one CCD, so it wouldn't be affected.

 

[cellarnoise]

The lack of Zen5 turin server stuff sucks... All you keep enjoying these toys 🙂

I want 64 core or more leaks, cause this is boring... HAHA

 

[Markfw]

The lack of Zen5 turin server stuff sucks... All you keep enjoying these toys 🙂

I want 64 core or more leaks, cause this is boring... HAHA

10 October, (If my leaks are correct) I will fire up my Turin and give you some info !

 

[Josh128]

AMD did say they will fix the core latency issue so hopefully more performance yet to be unlocked

When did they say this? Link??

 

[CakeMonster]

AMD did say they will fix the core latency issue so hopefully more performance yet to be unlocked

If true, that's great for the people getting 1 CCX cpu's. But it doesn't make the 9950X with core parking mechanisms and thread prioritization schemes more attractive than a 7950X.

 

[MS_AT]

If true, that's great for the people getting 1 CCX cpu's. But it doesn't make the 9950X with core parking mechanisms and thread prioritization schemes more attractive than a 7950X.

And did anyone check if core parking helps?🙂

 

[Mahboi]

AMD Ryzen 5 9600X & Ryzen 7 9700X Linux Performance With 105 Watt cTDP Review - Phoronix

www.phoronix.com

Aha, new TDP new power!
Actually no, it's really lame.
I'm surprised the extra power gives it so little. It's 50% more power for about 10% more perf. Absolutely not what I expected.

 

[positivedoppler]

When did they say this? Link??

https://wccftech.com/amd-to-provide...let Latency&text=AMD is reportedly working on,"Zen%205"%20Desktop%20CPUs.

 

[In2Photos]

When did they say this? Link??

Another article: https://www.notebookcheck.net/Horri...50X-is-reportedly-getting-fixed.880285.0.html

 

[Mahboi]

note to self: svgz images won't work.

I mean it looks stylish to see an 8 core Zen 5 beat Intel's best...but let's be honest nobody should care.
It's 50% more power draw on both the 6 and 8 cores. For a growth level of 10% at best.

 

[Philste]

I'm surprised the extra power gives it so little. It's 50% more power for about 10% more perf. Absolutely not what I expected.

But why tho? There were plenty PBO Benches at Release Day, which showed the same 10% Uplift too.

 

[Mahboi]

But why tho? There were plenty PBO Benches at Release Day, which showed the same 10% Uplift too.

Well I was suspecting that the 8 core was seriously starved and would show more than 10% after the boost.
Turns out I was very wrong.

 

[Rheingold]

I'm surprised the extra power gives it so little.

It's more than in the last generation. The 9700X gets +6.56% with the higher power limits, compared to just +4.14% between 7700 and 7700X.

 

[In2Photos]

The crazy thing is had AMD released the 9700X as a 105W chip and it had the additional 10% uplift, no one would have cared what the results were at 65W. But since it was released at 65W and reviewed that way, now all anyone will talk about is the cost of the performance increase, i.e. 40% more power for only 10% more performance.

 

[Josh128]

Another article: https://www.notebookcheck.net/Horri...50X-is-reportedly-getting-fixed.880285.0.html

Yeah, Im familiar with that, but that is not AMD saying they will fix it. Thats a rumor from some unknown Asian person on X.