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

[DisEnchantment]

Speculate at will

 

[poke01]

A lot of companies are using N3E this year, it’s an okay node.

Anyway Zen5 on client isn’t finished yet. Strix halo will be on N3E and will have much better bandwidth.

 

[Joe NYC]

Do you have an AMD press release stating current Turin processors are on N3E? Google AI is saying some of the Turin processors are on N3E (highest end chips) and others are on N4P. That's AI but I always thought all Turin processors would be on at least N3.

The classic (vanilla) Turin CCDs are on N4P, same as desktop. Turin Dense is on N3E.

In AMD press release, they only mentioned that these processors use N3 and N4, without specifying the specific iteration of the nodes.

 

[marees]

Do you have an AMD press release stating current Turin processors are on N3E? Google AI is saying some of the Turin processors are on N3E (highest end chips) and others are on N4P. That's AI but I always thought all Turin processors would be on at least N3.

There was exactly 1 reference I could get from googling this:

According to AMD, the CCDs on this chip were fabbed on a 3nm process (undoubtedly TSMC’s), with AMD apparently looking to take advantage of the densest process available in order to maximize the number of CPU cores they can place on a single chip.

https://www.anandtech.com/show/2142...-processors-up-to-192-cores-coming-in-h2-2024

There is also Tom's, but not sure how reliable they are:

The Zen 5c models employ up to 12 3nm CCDs with 16 Zen 5c cores per chiplet paired with the same I/O die.

https://www.tomshardware.com/pc-com...gen-zen-5-chips-with-up-to-192-cores-500w-tdp

 

[inquiss]

There was exactly 1 reference I could get from googling this:

According to AMD, the CCDs on this chip were fabbed on a 3nm process (undoubtedly TSMC’s), with AMD apparently looking to take advantage of the densest process available in order to maximize the number of CPU cores they can place on a single chip.

https://www.anandtech.com/show/2142...-processors-up-to-192-cores-coming-in-h2-2024

There is also Tom's, but not sure how reliable they are:

The Zen 5c models employ up to 12 3nm CCDs with 16 Zen 5c cores per chiplet paired with the same I/O die.

https://www.tomshardware.com/pc-com...gen-zen-5-chips-with-up-to-192-cores-500w-tdp

It's as others have stated. Dense is on n3e and desktop and standard Turin on n4p

 

[Josh128]

A lot of companies are using N3E this year, it’s an okay node.

Anyway Zen5 on client isn’t finished yet. Strix halo will be on N3E and will have much better bandwidth.

The CCDs wont be on N3(E?), just the GPUIO die.

 

[poke01]

The CCDs wont be on N3(E?), just the GPUIO die.

That sucks. At least there is better bandwidth to look forward too.

 

[Timmah!]

I have 8 Zen 4 desktops. It faster than those, takes less power and runs cooler, so in my opinion, even the desktop it is still a winner.

Each to his own. If you don't like it, buy a failed and degrading raptorlake.

Then again you are well aware you are niche within a niche with your 8 zen4 CPUs and same goes for your avx512 use case. Its only marginally faster outside of that specific use case and few other, mostly server-like loads, taking less power and running cooler is bit debatable as well. Bottom line, your perspective is different, and yeah, absolutely, to each their own. Not being enamored with Zen5 does not mean one has to be hasting to get Intel, in my case i will simply hold on my 7950x most likely. 9950x3D might still sway me, though, if the rumors/leak about clock parity with vanilla chips are true.

 

[RnR_au]

N3E is not yet available yet.

You may be thinking of N3P...

Building on the back of their current-generation N3E process, the optical shrink of this process technology, N3P, is now on track to enter mass production in the second half of 2024.

From https://www.anandtech.com/show/2139...nology-on-track-for-mass-production-this-year

 

[lightmanek]

Then again you are well aware you are niche within a niche with your 8 zen4 CPUs and same goes for your avx512 use case. Its only marginally faster outside of that specific use case and few other, mostly server-like loads, taking less power and running cooler is bit debatable as well. Bottom line, your perspective is different, and yeah, absolutely, to each their own. Not being enamored with Zen5 does not mean one has to be hasting to get Intel, in my case i will simply hold on my 7950x most likely. 9950x3D might still sway me, though, if the rumors/leak about clock parity with vanilla chips are true.

You are quite wrong on bolded part. There are quite a few workstation and productivity tasks where Zen5 is easily 10%-20% faster than Zen4. In some cases gains are very significant, like in Cinema4D viewport, where my new Ryzen 9950X scores significantly higher than Zen4 did (20%).

 

[OneEng2]

There will be a lot of moving targets. Yesterday, everyone absolutely needed Thin and Light with ability to watch 24 hours of Netflix non-stop.

When AMD releases Kraken, and it has similar efficiency, lower price, but slightly slower GPU, all these Netflix watchers will become gamers overnight..

LOL. Yep. Never knew you even needed it until you had it .

N3E is not yet available yet. Only Apple has N3E coming on their M4 chip later this year. I think Zen 5 Turin is running standard N3 but later versions are supposed to be on N3E. I could be wrong but Apple is 1st in line to get the cutting edge silicon from TSMC.

N3E (as others have pointed out) is out; however, you are correct that I mis-spoke. Intel processors are on N3B ..... which is more dense than N3E btw, just more expensive with lower yields I believe. N3P is being released by the end of this year I have heard.

AMD is using N3E for Turin.

This is what I had heard as well. Can anyone confirm that Turin dense is using N3E for the Zen 5c CCD's and not just the io die?

 

[OneEng2]

A lot of companies are using N3E this year, it’s an okay node.

Anyway Zen5 on client isn’t finished yet. Strix halo will be on N3E and will have much better bandwidth.

Hadn't heard that Strix halo would be on N3E. I also saw a post saying only the IO die would be N3E and that the Zen 5 cores would be on N3P. Seems like no one can agree on it yet.

 

[marees]

H

Can anyone confirm that Turin dense is using N3E for the Zen 5c CCD's and not just the io die?

How about this

According to AMD, the CCDs on this chip were fabbed on a 3nm process (undoubtedly TSMC’s), with AMD apparently looking to take advantage of the densest process available in order to maximize the number of CPU cores they can place on a single chip.

https://www.anandtech.com/show/2142...-processors-up-to-192-cores-coming-in-h2-2024

 

[Josh128]

LOL. Yep. Never knew you even needed it until you had it .

N3E (as others have pointed out) is out; however, you are correct that I mis-spoke. Intel processors are on N3B ..... which is more dense than N3E btw, just more expensive with lower yields I believe. N3P is being released by the end of this year I have heard.

This is what I had heard as well. Can anyone confirm that Turin dense is using N3E for the Zen 5c CCD's and not just the io die?

The Turin IOD's are 6nm.

 

[Joe NYC]

Hadn't heard that Strix halo would be on N3E. I also saw a post saying only the IO die would be N3E and that the Zen 5 cores would be on N3P. Seems like no one can agree on it yet.

Well, it will be a gigantic IO die (SoC). It will have a GPU (comparable to x60 ot x70 dGPUs), NPU, 32 MB of cache, maybe some low power cores, and then, of course IO.

Zen 5 cores attacked to this die will be close to standard desktop dies, but with slightly changed connectivity. More likely than not, they will be N4P based. If for no other reason, there are not any full Zen 5 on anything other than N4P.

 

[RnR_au]

Hadn't heard that Strix halo would be on N3E.

Its a rumour from MLID back in June and he said that only the IO die on Strix Halo would be N3E. A weird use of expensive silicon tbh.

 

[Tuna-Fish]

Its a rumour from MLID back in June and he said that only the IO die on Strix Halo would be N3E. A weird use of expensive silicon tbh.

The "IOD" on Strix Halo is also the GPU.

 

[RnR_au]

The "IOD" on Strix Halo is also the GPU.

Ahh ofcause... brain fart on my part.

 

[inquiss]

LOL. Yep. Never knew you even needed it until you had it .

N3E (as others have pointed out) is out; however, you are correct that I mis-spoke. Intel processors are on N3B ..... which is more dense than N3E btw, just more expensive with lower yields I believe. N3P is being released by the end of this year I have heard.

This is what I had heard as well. Can anyone confirm that Turin dense is using N3E for the Zen 5c CCD's and not just the io die?

Turin dense io die is 6nm. The cores are n3e.

 

[Timmah!]

You are quite wrong on bolded part. There are quite a few workstation and productivity tasks where Zen5 is easily 10%-20% faster than Zen4. In some cases gains are very significant, like in Cinema4D viewport, where my new Ryzen 9950X scores significantly higher than Zen4 did (20%).

The Cinema thing actually interests me, or better said, i wonder if same speed-up applies to 3dsmax. By "viewport" you mean more performance/FPS when zooming/camera panning around the modeling space, not refering to rendering speed, right?
Regarding me being wrong, i see it more as difference in what one considers as "significant". You say its easily 10-20 percent faster, i say thats for the most part not perceivable speed-up (10 percent definitely, 20 percent is borderline pushing it), so it does not justify dropping 700 EUROs on the new CPU - to me.

Like, for example, i am using 4090 for work with octane render, and the only upgrade path for me is 5090 (assuming its again about 2x as fast as 4090 for the particular task, like 4090 was compared to 3090 and 3090 compared to 2080Ti). Hypothetical 5080, thats about 10 percent faster than 4090, does nothing for me, if Nvidia wont sell 5090 at all and 5080 gonna be the new top card, i am not paying 1200 for it, fooling myself 10 percent of additional performance is "significant".

 

[PJVol]

AMD does advertise Max boost, 1 core clock speed. That's 5.0 GHz for 7800x3d. So we will see what it is for 9800x3d

So what? What exactly in my post you don't agree with?

 

[Josh128]

So new Chiphell leak about 9800X3D.



"Frequency is quite high". lol OK. That should rule out anything under 5.4GHz, as the entire Zen 5 desktop lineup bottoms at 5.4GHz. Lower than the lowest is not "quite high". I still maintain that listed frequency iso with 9700X is impossible, so if this guy is for real, then its about a 98% chance 5.4GHz "max boost" will be on the box.

Why the leaker is suggesting to wait for 9950X3D due to 9800X3D pricing is beyond my imagination though. We've seen the MSI leaks showing less fps for the 9950X3D than the 9800X3D-- why they suggest one wait for that because 9800X3D will be expensive when 9950X3D will obviously be even more expensive and slower in gaming??

 

[CouncilorIrissa]

The dual-CCD SKUs should behave exactly the same as single-CCD ones due to X3D driver straight up disabling the other CCD.

 

[CakeMonster]

What if you do leave other medium to heavy CPU demanding tasks running in the background (encoding, AI generating, etc) but want to play a game at the same time? Will the secondary CCD chug along and the primary CCD run a lot slower? Will it run worse or better than the CPU's that don't have that driver like the 7950X?

 

[CouncilorIrissa]

What if you do leave other medium to heavy CPU demanding tasks running in the background (encoding, AI generating, etc) but want to play a game at the same time? Will the secondary CCD chug along and the primary CCD run a lot slower? Will it run worse or better than the CPU's that don't have that driver like the 7950X?

I don't have one, but from what I've read the second CCD would wake up as soon as you alt-tab out of the game.

 

[lightmanek]

The Cinema thing actually interests me, or better said, i wonder if same speed-up applies to 3dsmax. By "viewport" you mean more performance/FPS when zooming/camera panning around the modeling space, not refering to rendering speed, right?
Regarding me being wrong, i see it more as difference in what one considers as "significant". You say its easily 10-20 percent faster, i say thats for the most part not perceivable speed-up (10 percent definitely, 20 percent is borderline pushing it), so it does not justify dropping 700 EUROs on the new CPU - to me.

Like, for example, i am using 4090 for work with octane render, and the only upgrade path for me is 5090 (assuming its again about 2x as fast as 4090 for the particular task, like 4090 was compared to 3090 and 3090 compared to 2080Ti). Hypothetical 5080, thats about 10 percent faster than 4090, does nothing for me, if Nvidia wont sell 5090 at all and 5080 gonna be the new top card, i am not paying 1200 for it, fooling myself 10 percent of additional performance is "significant".

Yes, it is live preview in 3D


The car chase scene in case of Cinebench, but similar gains were measured in Solidworks.

Also, wireframe and shaded viewports from older software gained massively. I only have Zen 3 comparison on this table, but the jump is almost unbelievable for this older code.


I understand your point, but for most people willing to upgrade from AM5, they do have a CPU they can sell, so it's not 700Euro they invest. In my case, I sold my 7800X3D for £330 and paid £570 for new CPU, it costed me £240 to go from 8 cores to 16 cores. I lost some performance in games, but gained a lot everywhere else. Besides, I don't feel any real negative impact on my gaming as I play at 1440p and maxed out. There are no games which are unplayable on Ryzen 9950X, but for productivity I save gobs of time on anything taking advantage of extra cores and a bit of time on every other app mostly singe-threaded.