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

[DisEnchantment]

Speculate at will

 

[static shock]

AIs are tested on all kind of forums...

I go with the before mentioned leak >40% normalized clockspeed performance in this forum.
Man! "zen5" use a execution engine of 6 Alus 50% more Plus 50% more fpus Floating point units.
Congratulations for the score of team red!

 

[static shock]

That was cancelled 16 months ago

Problem is that RDNA3.5! Rdna4 whoever know when it lands!
AMD must go after mindshare figthing! Just going TSMC 3np instead of TSMC 4np with mega efficient videocards won't help. Enthusiasts are crazy to buy the "bestest" card, but going over with the efficient cards do a little help if they both outperform Sarlak gpu by two fold at least.

 

[static shock]

Just one thing, KeplerL2. Moore Law is dead is not a great leaker for what we know. Original Zen4 was supposed to bring 29% more ipc. And Morpheus was supposed to be a 2024 launch.
Zen5 embraces the empowered 30% previous tenstorrent slide plus another year of work(2023).

 

[DrMrLordX]

And Morpheus was supposed to be a 2024 launch.

Morpheus is Zen6. It was never 2024.

 

[poke01]

I would say Cinebench 2024 is a good benchmark. It correlates pretty well with Blender now.

the extra e cores are now normalised and it’s no longer an e core spam machine like Cinebench R23 was.

The AMD 7950X is a rendering beast due to having 16 P cores. The fact that Intel scores lowers cause those e cores kinda suck for heavy tasks.
AMD will likely present a Blender benchmark nxt week and we can infer how fast it compares to the their current CPUs below.

Desktop:



Laptop:

 

[adroc_thurston]

I would say Cinebench 2024 is a good benchmark. It correlates pretty well with Blender now.

It's still a pile of rather antique SSE.

 

[poke01]

It's still a pile of rather antique SSE.

Agree but much better than r23.

 

[adroc_thurston]

Agree but much better than r23.

Kind of.
Blender is still better since underlying kernels are way more modern.

 

[Timmah!]

Kind of.
Blender is still better since underlying kernels are way more modern.

In real world, for actual work, neither is the top choice. Its Vray or Corona.

 

[igor_kavinski]

Its Vray or Corona.

Is Blender slower or missing some key features? I believe Blender would be chosen in instances where the budget is really tight.

 

[Timmah!]

Is Blender slower or missing some key features? I believe Blender would be chosen in instances where the budget is really tight.

Well, surely people who do this stuff for something not really important to them, would pick Blender, as its free. But people who do it professionally, stick to Vray or Corona, as they are older and more feature complete. And one important thing, if you check sites like 3Dsky, CGtrader, etc... where you can buy pre-made 3D models of ton of stuff, they are generally set-up with Vray or Corona materials. Means you just buy them, download them, put them into your scene and render. No material translation required, that can be pain and annoying, i shall know, since i use Octane Render myself, and have to do that a lot.

I can see GPU renderers being the future, and especially stuff like Unreal, Lumion and D5 taking over this field, but it will take years. If you are used to certain workflow and it works for you, even if there are newer and better ways to do it, people gonna stick to what they know and its enough to them. At least in my personal experience.

 

[igor_kavinski]

since i use Octane Render myself

For material translation? By the way, any of your projects appear in a Hollywood movie? Just asking

 

[static shock]

Zen4 was originally going to deliver 29% more IPC than Zen3.

EDIT: Zen5 is most likely to be faster than Arrow Lake beinging 49.5% more IPC over Zen4.

 

[igor_kavinski]

EDIT: Zen5 is most likely to be faster than Arrow Lake beinging 49.5% more IPC over Zen4.

Oh no, that's a fail. Why couldn't they manage a measly 0.5% more?

 

[static shock]

Can't say.

 

[igor_kavinski]

Can't say.

You can say in vague terms. Like using plush toy animals as analogies. We just need hints.

 

[static shock]

Zen 5 was supposed to bring 30% more IPC over Zen4. Then Zen6 was supposed to bring 10-15% more IPC over Zen5. Got it?

 

[Kaffeekenan]

Can't say.

First I wanted to ignore you. Now you get annoying to me and I will block you...just leave the forum, we do not need spammers here who contribute absolutely nothing meaningful. Everyone here knows, that you know nothing about Zen5 so you can stuff your "can't say" and "can't say my source" and stuff like that in your bag and just leave.

 

[igor_kavinski]

Zen 5 was supposed to bring 30% more IPC over Zen4. Then Zen6 was supposed to bring 10-15% more IPC over Zen5. Got it?

But what happened? Why did they go for 19.5% additional IPC and not 20%? That's a weird number. Almost like someone on the team is superstitious or something.

 

[FlameTail]

Zen 5 was supposed to bring 30% more IPC over Zen4. Then Zen6 was supposed to bring 10-15% more IPC over Zen5. Got it?

Is that from the MLID leak?

 

[static shock]

The Original ZEN5 was the "original" Nirvana. "Morpheus" was the "Original" Zen6.

Arrow Lake will be likely just a bit slower than Zen5.

 

[static shock]

Is that from the MLID leak?

He's not a good source with all outdated slides. The only thing he was right was that Zen5 would be greater and wider than Zen4.

 

[igor_kavinski]

The Original ZEN5 was the "original" Nirvana. "Morpheus" was the "Original" Zen6.

Arrow Lake will be likely just a bit slower than Zen5.

So you have no idea about ARL.

And you broke your NDA by revealing both Nirvana and Morpheus original designs have been compromised with "as yet unknown" changes. Pack your bags, mate. Time to disappear!

 

[static shock]

So you have no idea about ARL.

It is your tought.

 

[igor_kavinski]

It is your tought.

Then why you said "likely"?