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

[DisEnchantment]

Speculate at will

 

[TESKATLIPOKA]

they claim its result at 35w, dont they? so it does score 41 percent better than previous gen 65w cpu, while not even having 12 of those new full fat cores like that one did? could zen5c cores be significantly more performant than regular zen4 ones?

i guess i hope its true. This could actually mean that supposed 2400 score for the 16c is some bogus and its way more, cause no way could both of these be true.

at 40 percent more MT performance, it would be actually interesting proposition.

btw that 2185 score thats being refered for 7950x in cb24 - would love to know how they achieved that. I got 2120 on mine, and i tuned the cpu to run at 5,212 avg all core, which is about 150 mhz more than it was at default out of the box, and cooling it with arctic 420, so pretty much the best non-custom possible solution. i mean, 65 more points, thats beyond that usual result fluctuation.

You managed only 2120 points with 16C32T Zen4 running at 5.2GHz.
For this Strix Point to manage 2000 points It would need +33% IPC and 4.9GHz.
I don't think It's possible at 35W. Even at 135W It would be a great result.

Not addressing the video I didn’t watch, but as to your claim that much higher scores for nT workloads “is impossible “, that is not true.

Multicore workloads in Intel/AMD chips are almost always power limited. Reducing power consumption can lead to healthy performance gains IF the scalability is there and there are no bottlenecks.

I am not saying this is the case here, but it is a possibility.

@Timmah! managed only 2120 pts with 7950x, which was running at 5.2GHz.
7900 at 5.2GHz should be ~1590 pts or a bit more.
Strix Point to manage 2400 pts would need:
A) ~33% higher IPC than Zen4 and 5.9GHz
B) ~50% higher IPC than Zen4 and 5.25GHz
And this is a mobile chip where most of the cores are Zen5c, so less power efficient than the standard Zen5 at higher clocks.
So I stand by what I originally said.

BTW, It turns out It was "only" 2000 pts.

 

[Timmah!]

Anyway, if this is supposed to be late ES, close to eventual retail samples, could the release actually happen at CES? I mean, i was expecting it more like Computex time, Q2 next year, not right away....based on what was claimed previously.
While i am very curious about the performance, the later date would suit me more, as i bought the 7950x last December and only built the rig like in late April - at last. So i am not quite in a "mood" to replace it not even a year later, but if its gonna be good, i would want to, i know myself. Additionally, just forked 730 EUROs for new iPhone. Damn :-D

BTW, while we still talk about this as about 8950x/8950HS, etc... it will almost certainly be called 9000 series

 

[igor_kavinski]

it will almost certainly be called 9000 series

9 is for 2025 products, if AMD stays true to their naming scheme.



But all bets are off if their marketing dept thinks we are getting 2025 performance in 2024...

 

[adroc_thurston]

BTW, while we still talk about this as about 8950x/8950HS, etc... it will almost certainly be called 9000 series

no

9 is for 2025 products, if AMD stays true to their naming scheme.

yes

 

[Joe NYC]

no

So Strix Halo are going to be 2024 models? Still planned for 2024?

 

[Timmah!]

no

yes

oh, forgot about this utter lunacy.
my bad then

 

[linkgoron]

oh, forgot about this utter lunacy.
my bad then

I'm hopeful that it'll be thrown out just like that old Radeon naming scheme that never went anywhere.

 

[Timorous]

I'm hopeful that it'll be thrown out just like that old Radeon naming scheme that never went anywhere.

View attachment 88418

Nah.

The mobile naming scheme is good because you actually know what version of Zen is in a device. Different form factors and price ranges mean different Zens are a best fit so knowing a device is Zen 3 or Zen 4 immediately allows the user to make an informed decision.

 

[Thibsie]

Nah.

The mobile naming scheme is good because you actually know what version of Zen is in a device. Different form factors and price ranges mean different Zens are a best fit so knowing a device is Zen 3 or Zen 4 immediately allows the user to make an informed decision.

I agree. I just looked at Lenovo last ThinkPads and they now come with 7xxx range but some are 7x3x others are 7x4x. At least, that part is clear.

 

[linkgoron]

I agree. I just looked at Lenovo last ThinkPads and they now come with 7xxx range but some are 7x3x others are 7x4x. At least, that part is clear.

For informed users it was always pretty simple to differentiate (e.g. ryzen 5 gen had odd as zen 2 and even as zen 3). It's going to be very confusing once 8 parts start getting released, for the average consumer.

 

[Timmah!]

I have no issue with setting up a new, clear system for improved transparency of the product lines and SKUs. Thats actually commendable, if it works. But wheh you do that, you shall restart entire naming scheme, not adjust existing one, that functioned in different manner for past 3 or more generations.
So they went from 3000 to 5000 to 7000 generations, but now its gonna be 8000 just because. It hurts my OCD

 

[Joe NYC]

I have no issue with setting up a new, clear system for improved transparency of the product lines and SKUs. Thats actually commendable, if it works. But wheh you do that, you shall restart entire naming scheme, not adjust existing one, that functioned in different manner for past 3 or more generations.
So they went from 3000 to 5000 to 7000 generations, but now its gonna be 8000 just because. It hurts my OCD

Rembrandt was 6000 generation.

 

[moinmoin]

So they went from 3000 to 5000 to 7000 generations

If you wanted to say AMD skipped 4000 and 6000, they didn't.

 

[igor_kavinski]

If you wanted to say AMD skipped 4000 and 6000, they didn't.

They skipped 6000 on desktop.

 

[moinmoin]

They skipped 6000 on desktop.

I wouldn't word it that way. They didn't skip it. They just didn't use it on desktop (for various reasons).

 

[Timorous]

For informed users it was always pretty simple to differentiate (e.g. ryzen 5 gen had odd as zen 2 and even as zen 3). It's going to be very confusing once 8 parts start getting released, for the average consumer.

not really.

8x20 may or may not exist. but if it does it will be Zen 2.
8x30 will be Zen 3
8x40 will be Zen 4
8x50 will be Zen 5

RDNA2 vs RDNA3 vs RDNA 3.5 iGPUs will be 600 or 700 or 800 on the GPU spec so that is also pretty clear.

The biggest issue with be something like an 8740 vs an 8550 in the same product range but really it just means the former has an nT advantage and the latter a sT advantage unless AMD totally bork the Ryzen 3,5,7,9 branding, which is always possible. The other option is that there won't be an 8740 and 8550 within the same product range as they target different segments so that cross shop may not even be especially valid.

 

[igor_kavinski]

They just didn't use it on desktop (for various reasons).

It's still a very odd omission. Until and unless they do otherwise, I'm forced to think that there won't be a desktop AM5 successor to 5600G/5700G.

 

[TESKATLIPOKA]

I wonder what will happen to 2026 products. That should use 10
Will they be called Ryzen 7 10840HS or Ryzen 9 10845HX for example?
Or they will use Ryzen 7 X840HS or Ryzen 9 X845HX?

 

[coercitiv]

So they went from 3000 to 5000 to 7000 generations, but now its gonna be 8000 just because. It hurts my OCD

It will probably happen again every few years, mobile has yearly product updates (even if just rebranding) while desktop will update every 1.5 years at best. So they'll skip a desktop generation every now and then to keep them in sync.

 

[moinmoin]

It's still a very odd omission. Until and unless they do otherwise, I'm forced to think that there won't be a desktop AM5 successor to 5600G/5700G.

They "skipped" 4000 on desktop as well until it was released much much later (probably why Timmah! didn't think of it). And that can also still happen with 6000, though I don't expect it at this point.

 

[igor_kavinski]

I wonder if adroc can tell us how many dies AMD has to accumulate before releasing a new SKU.

 

[Timmah!]

I wouldn't word it that way. They didn't skip it. They just didn't use it on desktop (for various reasons).

And now they are apparently using it, breaking the scheme. Thats my issue. I do understand they want to do it a new way, but then they should have make it significantly different from the old system, not hybridize them.
I admit its the least important thing.

 

[Timmah!]

not really.

8x20 may or may not exist. but if it does it will be Zen 2.
8x30 will be Zen 3
8x40 will be Zen 4
8x50 will be Zen 5

RDNA2 vs RDNA3 vs RDNA 3.5 iGPUs will be 600 or 700 or 800 on the GPU spec so that is also pretty clear.

The biggest issue with be something like an 8740 vs an 8550 in the same product range but really it just means the former has an nT advantage and the latter a sT advantage unless AMD totally bork the Ryzen 3,5,7,9 branding, which is always possible. The other option is that there won't be an 8740 and 8550 within the same product range as they target different segments so that cross shop may not even be especially valid.

Wait, the 5 in the 8950 means Zen5 now? And 9 i guess means the ryzen 5,7,9 moniker, right?
If yes, how do you differentiate between 12C and 16C that both used 9? I guess by the last of the four numbers - does it mean that replacement of 7950x will be say 8951x and replacement of 7900x will be say 8952x?

 

[TESKATLIPOKA]

Wait, the 5 in the 8950 means Zen5 now? And 9 i guess means the ryzen 5,7,9 moniker, right?

Don't think so. Something like that.
Phoenix is for example
Ryzen 7 7840H
Ryzen 7 7840HS
Ryzen 7 7840U
That 9 should mean for what market It is intended or something like that.

edit: Here It is

 

[Joe NYC]

I wonder what will happen to 2026 products. That should use 10
Will they be called Ryzen 7 10840HS or Ryzen 9 10845HX for example?
Or they will use Ryzen 7 X840HS or Ryzen 9 X845HX?

They can go hexadecimal