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

[DisEnchantment]

Speculate at will

 

[HurleyBird]

I hope performance, and i honestly hope Adroc is truthful and its higher, like 50 percent

I hope so too, of course. But if I had to put money on it either way, it would have to go on the side of the hype train derailing, as it almost always does.

 

[Ajay]

I hope performance, and i honestly hope Adroc is truthful and its higher, like 50 percent 😛 Even if its 50 percent IPC

I consulted my toaster oven and it assured me that the IPC gain would be > 50% with no frequency regression. My toaster oven has impeccable sources, or so it says.

 

[HurleyBird]

I hope so too, of course. But if I had to put money on it either way, it would have to go on the side of the hype train derailing, as it almost always does.

Actually, let me expand on this. The odds of any hype train being realized in an environment where rather unexciting internal projections are dismissed as sandbagging and hyper-conservatism, and where whispers of development problems and missed targets are starting to materialize, is effectively zero.

It would be nice to be wrong on this one, but I'm sure I'm not.

 

[RnR_au]

I consulted my toaster oven and it assured me that the IPC gain would be > 50% with no frequency regression. My toaster oven has impeccable sources, or so it says.

Well your AI is better than mine. ChatGPT 3.5, via OpenAI won't even hazard a guess for me. It starts whining about its training. I asked for estimates but it still refused to give a number. It seems AI is even more cagey than Youtube 'leakers' 😛

 

[Abwx]

After RGT spilled thoses speculative beans i m eagerly awaiting for some MLID s devastating counter attack with due numbers from unautorised sources, the best leaker title and supremacy is at play here.

 

[FlameTail]

After RGT spilled thoses speculative beans i m eagerly awaiting for some MLID s devastating counter attack with due numbers from unautorised sources, the best leaker title and supremacy is at play here.

Idk but I feel MLID is more legit than RGT is.

 

[FlameTail]

List of World's Fastest Trains:

10. Korail KTX-Sancheon: 305km/h (Korea)
9. Renfe AVE 103: 310km/h (Spain)
8. ONCF Al Boraq: 320km/h (Morocco)
7. JR Shinkansen: 320km/h (Japan)
6. SCNCF TGV: 320km/h (France)
5. DB ICE: 350km/h (Germany)
4. CR Fuxing: 350km/h (China)
3. CR Harmony: 350km/h (China)
2. Shanghai Maglev: 460km/h (China)
1. Zen 5 Hype Train (AMD)

 

[adroc_thurston]

Idk but I feel MLID is more legit than RGT is.

Neither of them have any idea what they're talking about.
The former was sperging about ACM+ not existing on twitter like a week ago.

 

[Gideon]

Maybe It’s a journey?

https://x.com/olrak29_/status/1753210909395489259?s=46&t=tNnMS9VB0KADxuy7OaVjLQ

 

[adroc_thurston]

Maybe It’s a journey?

https://x.com/olrak29_/status/1753210909395489259?s=46&t=tNnMS9VB0KADxuy7OaVjLQ

It's Zen5% so it gotta be 5%.
Trust the plan.

 

[gdansk]

There we go. Now we can't disappointed if we're expecting -5% IPC increase.

 

[Fjodor2001]

There we go. Now we can't disappointed if we're expecting -5% IPC increase.

We can still be disappointed. It can turn out to be -10% when released. 🤣

 

[coercitiv]

Maybe It’s a journey?

More than a journey, a sacrifice. In order to wield Zen 5, one must let go of those dearest to their heart.



And so will come the day of the first Zen 5 wielder, when half of the leakers will vanish in oblivion. Perfect balance, restored.

 

[Tuna-Fish]

Idk but I feel MLID is more legit than RGT is.

It's hard to judge. Both of them have definitely leaked material that was not public before. Both of them have also sometimes completely flubbed analysis of what they had.

 

[Timorous]

Idk but I feel MLID is more legit than RGT is.

They both flail in the dark but MLID is a better presenter.

 

[FlameTail]

They both flail in the dark but MLID is a better presenter.

That's reflected in their subscriber count I guess.

 

[CouncilorIrissa]

MLID's self-congratulatory antics is much more annoying, however.

 

[eek2121]

I only had in mind the expansion of the core logic and the resulting IPC gain.It is known that Yonah's biggest problem was the lower maximum clock speed, so Conroe's much higher IPC and achieving a much higher clock speed resulted in a significant increase in Core 2 performance.

I remember my jaw dropping when Conroe launched and I saw the benchmarks in comparison to the clock speeds. It made the Pentium 4 obsolete.

We desperately need another Conroe style correction in that regard. IMO Too much silicon (and power, for that matter) is wasted chasing high clocks.

Yonah was mobile only, so that was likely the reason you never saw higher clocks.

Conroe was also super successful because that was around the time Apple switched to x86.

Hopefully AMD gets things going in the right direction with Zen 5.

 

[FlameTail]

What is the average ST performance improvement YoY for CPU cores?

Considering all the main players: Intel, AMD, Apple, ARM

 

[SteinFG]

What is the average ST performance improvement YoY for CPU cores?

Considering all the main players: Intel, AMD, Apple, ARM

You should watch just a couple reviews and come to the ballpark conclusion.

 

[Thibsie]

I remember my jaw dropping when Conroe launched and I saw the benchmarks in comparison to the clock speeds. It made the Pentium 4 obsolete.

We desperately need another Conroe style correction in that regard. IMO Too much silicon (and power, for that matter) is wasted chasing high clocks.

Yonah was mobile only, so that was likely the reason you never saw higher clocks.

Conroe was also super successful because that was around the time Apple switched to x86.

Hopefully AMD gets things going in the right direction with Zen 5.

Yonah was also 32bit, alas.
I still have an old Thinkpad X60.
A pity 'cos with 8 GB ram, it would have been useful a lot longer.

 

[AMDK11]

I remember my jaw dropping when Conroe launched and I saw the benchmarks in comparison to the clock speeds. It made the Pentium 4 obsolete.

We desperately need another Conroe style correction in that regard. IMO Too much silicon (and power, for that matter) is wasted chasing high clocks.

Yonah was mobile only, so that was likely the reason you never saw higher clocks.

Conroe was also super successful because that was around the time Apple switched to x86.

Hopefully AMD gets things going in the right direction with Zen 5.

But Netbrust was a high-clock microarchitecture with a 1-way x86 decoder.

Today, GoldenCove has a 6-way decoder that is wider than previous generations, unlike Netbrust whose decoder was narrower. The Pentium III had a 3-way decoder.

So the massive growth of IPC Conroe was no miracle compared to Netbrust. Compared to IPC Yonah, it was simply the generational leap that could be expected from the next generation. Additionally, the effect was intensified by the slightly exaggerated drama that Intel would no longer be able to develop a new architecture with a higher IPC than the K8.

 

[FlameTail]

You should watch just a couple reviews and come to the ballpark conclusion.

15% ?

 

[Glo.]

Zen 4 based 8700G scores around 2720-2750 pts in Geekbench 6.

If Zen 5 brings 30% IPC increase - we are looking at 3550 score in GB6 for Strix Point(low clocked, relatively).
If Zen 5 brings 40% IPC increase - we are looking at 3850 pts in GB6 for Strix Point.

If Zen 5 brings 5% decrease in IPC we are looking at 2600 pts in GB6 for Strix Point. Which would bring it to the level of Meteor Lake CPUs. 😛

In any case it will be interesting to watch the meltdown.

😛

 

[adroc_thurston]

But Netbrust was a high-clock microarchitecture with a 1-way x86 decoder.

Today, GoldenCove has a 6-way decoder that is wider than previous generations, unlike Netbrust whose decoder was narrower. The Pentium III had a 3-way decoder.

Decode is whatever tier stuff.
Netburst had a huge trace cache for that.
But it had so many glass jaws otherwise that the resulting IPC was kinda bad on a good day.
On a bad day, you were Tulsa and you were stuck in a replay loop.