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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

According to this Anandtech review:

AMD Zen 4 Ryzen 9 7950X and Ryzen 5 7600X Review: Retaking The High-End

www.anandtech.com

The average SPECINT 2017 ST increase from Zen3 to Zen4 is 23% considering the average of all substests, considering the platform differences there is probably some lower margin due to the cores only because of memory BW advantage of Zen4, which should not exist with Zen5.

Zen3 is the more relevant one since both are pure tocks.

 

[inf64]

According to this Anandtech review:

AMD Zen 4 Ryzen 9 7950X and Ryzen 5 7600X Review: Retaking The High-End

www.anandtech.com

The average SPECINT 2017 ST increase from Zen3 to Zen4 is 23% considering the average of all substests, considering the platform differences there is probably some lower margin due to the cores only because of memory BW advantage of Zen4, which should not exist with Zen5.

You have to adjust for the difference in clocks for both Zen 3 and Zen 4. Once adjusted, the difference is about 12-13% in favor of Zen 4 (per core, per clock, average of all subtests).
For Zen 2 vs Zen 3, the average is exactly 19%, just as AMD claimed.

 

[AMDK11]

SPEC2017 Zen3:

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

 

[inf64]

Zen 3 vs Zen 2 IPC in INT 1-thread as per AT : 1.514/ 1.264 = 1.197 or ~19%.
AMD claimed ~19% IPC improvement in their official docs.

 

[AMDK11]

So the +40% for Zen 5 in SPEC2017 is from Jim Keller's slide?

What impact can AVX512 have on the bottom line in SPEC?

 

[leoneazzurro]

I said

You have to adjust for the difference in clocks for both Zen 3 and Zen 4. Once adjusted, the difference is about 12-13% in favor of Zen 4 (per core, per clock, average of all subtests).
For Zen 2 vs Zen 3, the average is exactly 19%, just as AMD claimed.

I said "ST", not "IPC". In any case, 40%+, if confirmed, appears huge considering the Zen3->Zen4 jump, even if it factored further clock increases (and again, I am not saying there will be, only comparing generational jumps).

 

[adroc_thurston]

So the +40% for Zen 5 in SPEC2017 is from Jim Keller's slide?

no it's from Granite Ridge samples.

What impact can AVX512 have on the bottom line in SPEC?

see Zen4.

In any case, 40%+, if confirmed, appears huge considering the Zen3->Zen4 jump

In AMD history, only K7 was a bigger single-gen core iteration.
Expected.

 

[Abwx]

Dunno if it was already posted but that clear the things about the time line :

AMD added that although Strix Point will be released at the end of this year, it is expected that consumers will actually receive the product in the early first half of next year.

AMD Strix Point mit XDNA 2: Vorstellung in diesem Jahr, aber Verfügbarkeit erst 2025

Ein Termin für Strix Point im Jahr 2024 galt schon immer als ambitioniert, nun wird klar, dass AMD ähnlich dem Hawk-Point-Release verfährt.

www.computerbase.de

 

[adroc_thurston]

Dunno if it was already posted but that clear the things about the time line :

Faster but you're not getting Zen5 in mainstream devices (cheaper thinkpads and anything $800 SRP) until KRK1 launches.
Which is next year!

 

[AMDK11]

Is the SPEC2017 Granite Ridge sample test available online?

As for the K7, the core without L2 consists of 22 million transistors, while the Pentium III only has 9.5 million transistors. So considering that the K7 has over twice as many transistors compared to the Pentium III, it wasn't crazy. I realize that the lion's share of K7 is the huge for those times L1-I+L1-D 128KB compared to L1-I+L1-D 32KB in Pentium III.

It seems that L1 in K7 has quite an influence on the advantage of IPC over PIII.

 

[adroc_thurston]

Is the SPEC2017 Granite Ridge sample test available online?

Of course no, people would get sent into the dungeon for leaks of that caliber.

As for the K7, the core without L2 consists of 22 million transistors, while the Pentium III only has 9.5 million transistors

Good news everyone!
Zen5 is also loaded on logic xtors facilitating cool uarch features.

So considering that the K7 has over twice as many transistors compared to the Pentium III, it wasn't crazy.

Intel had a lot better L2 in all ways, bigger L1 on K7 was a cope measure more or less.

 

[AMDK11]

K7 core complexity compared to PIII over 2x xtors with only +6% higher IPC Integer. In FP there was a much bigger difference in favor of K7, which I cannot dispute.

 

[Kolifloro]

Then, I guess ... 'Strix Halo' on the shelves around autumn'25 @3NP ...

 

[adroc_thurston]

K7 core complexity compared to PIII over 2x xtors with only +6% higher IPC Integer.

Yes because it has to cope with a much worse L2.
Which also made the OG 1GHz Athlon a meme.

Then, I guess ... 'Strix Halo' on the shelves around autum'25 @3NP ...

-halo is CES'25.
If you look at the 2024 Zephy G14 it looks kinda sus with lower system power target over the 2022/2023 one.
You can guess why yourself.

 

[Kolifloro]

-halo is CES'25.
If you look at the 2024 Zephy G14 it looks kinda sus with lower system power target over the 2022/2023 one.
You can guess why yourself.

Honestly, nothing against of 'Zephyrus G14' ... but I tend to fantasize with a 'Stix Halo' mounted over a 'Thinkpad P16' ... ... that's why I am guessing it might happen around AUTUMN ...

Thank you either way !!!




Ps.- I also hope the 'delay' might help with the 3NP manufactoring process ...

 

[adroc_thurston]

but I tend to fantasize with a 'Stix Halo' mounted over a 'Thinkpad P16' ... ... that's why I am guessing it might happen around AUTUMN ...

Thinkpad updates are not really ever H2.

Ps.- I also hope the 'delay' might help with the 3NP manufactoring process ...

N3p is Zen6 stuff.

 

[AMDK11]

Yes because it has to cope with a much worse L2.
Which also made the OG 1GHz Athlon a meme.

View Pentium III Coppermine and K7 Athlon Thunderbird test results. Both have L2 256KB on the same chip as the core.


The differences are not big at all. Sometimes they are close to each other, sometimes one leads, sometimes the other, but there is not a huge/crushing gap between them. I doubt that the differences in favor of the K7 will be noticeable at all under normal use. Of course, depending on the application.

For my part, I'm finishing the K7 topic. Let's focus on Zen 5.

 

[Tuna-Fish]

Thunderbird was released many months after Coppermine, the lineup in the race to 1 GHz was Athlon Classic vs Coppermine.

In any case, 64/64 L1 caches were not a cope measure, they were genuinely very useful, especially for games. And when AMD fixed their L2 with thunderbird, that deficiency was gone while they still had the advantage of the bigger but just as fast L1.

 

[AMDK11]

Release time is one thing. For me it's about the technical aspect. If the average performance between Thunderbird and Coppermine is quite equal, the K7 on Slot-A must have performed worse because Thunderbird has about 8% advantage over Pluto.

 

[gdansk]

Regarding K7/Pentium 3 era:

The Katmai contains 9.5 million transistors, not including the 512 Kbytes L2 cache (which adds 25 million transistors).

With respect to Zen 5 I don't think Intel is in a good position. AMD can *affordably* spam cache to be competitive even if Arrow Lake is better than Zen 5 in 1T. And yet most signs point to Zen 5 being a much bigger change than Arrow Lake.

 

[adroc_thurston]

And yet most signs point to Zen 5 being a much bigger change than Arrow Lake.

It's very comical levels of that.
Gaming situation especially.

 

[Kaffeekenan]

So, b

It's very comical levels of that.
Gaming situation especially.
[/QUOTE

It's very comical levels of that.
Gaming situation especially.

Sooo, better than arrow lake in gaming? Are you sure?

 

[adroc_thurston]

So, b

Yea.
Major impacts on 2025 gaming laptop market.

 

[Gideon]

BTW, what happened @adroc_thurston and @Kepler_L2 that caused you to change your predictions ~10% higher than just a few weeks ago?:

ARL-S and Zen 5 Desktop Speculation

RULE RULE This spreadsheet is unmoderated. No spamming and please respect what others have written. I will stop making these public speculation spreadsheets once someone breaks the rule. That's all, have fun. (By the way, the tabs are at the bottom.)

docs.google.com

That google sheet was put up on March 9:

https://twitter.com/x/status/1766467520729375119

Zen 5 Hype Train taking off from the station?

 

[adroc_thurston]

BTW, what happened @adroc_thurston and @Kepler_L2 that caused you to change your predictions ~10% higher than just a few weeks ago?:

I never wrote anything in that chart so idk.
And yeah Granite Ridge SDVs went out into the wild.