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

[DisEnchantment]

Speculate at will

 

[fastandfurious6]

it's just scheduler shenanigans really, can't be otherwise

amd really lost an opportunity to do 2 things

1. 9950X3D with double Vcache - in each CCD

2. 9800HX3D 8 core 1ccd, mobile equivalent of 9800X3D & big market availability

if they did these two things they'd have complete market takeover in both top-end desktop and laptops

 

[511]

it's just scheduler shenanigans really, can't be otherwise

amd really lost an opportunity to do 2 things

1. 9950X3D with double Vcache - in each CCD

2. 9800HX3D 8 core 1ccd, mobile equivalent of 9800X3D & big market availability

if they did these two things they'd have complete market takeover in both top-end desktop and laptops

only if they have the volumes

 

[gdansk]

Extrapolating from comparisons of two different test systems to a moving target (285K) is foolish. I'd call it a waste of time but we have plenty of that and no other information yet.

 

[511]

Extrapolating from comparisons of two different test systems to a moving target (285K) is foolish. I'd call it a waste of time but we have plenty of that and no other information yet.

please tell me AMD didn't use Beta BIOS on ARL 🤣

 

[gdansk]

please tell me AMD didn't use Beta BIOS on ARL 🤣

Who knows what the configuration is. But Cyberpunk, which shows the largest difference between 9800X3D and 9950X3D via this proxy comparison, had a patch in the interim which improved ARL-S performance. That is strong evidence that this comparison is pretty much pointless.

 

[gaav87]

AMD CES IS LIVE:

 

[Thibsie]

please tell me AMD didn't use Beta BIOS on ARL 🤣

Intel.would totally have done so. I wouldn't be surprised if AMD did it too. No, that's not an excuse.

 

[desrever]

AMD doesn't seem to understand that 240% average is 2.4x performance. People who make AMD slides are so clueless its insane.

 

[gdansk]

140% faster is also a correct way to refer to 2.4x isn't it?

 

[Hitman928]

1.4x faster is also a correct way to refer to 2.4x isn't it?

1.4x = 40% faster.

2.4X = 140% faster.

 

[gdansk]

1.4x = 40% faster.

2.4X = 140% faster.

And they said 1.4x faster which is 140% faster unless I'm bad at percentages. So it seems technically correct but unusual wording.

 

[Tup3x]

View attachment 114349

AMD doesn't seem to understand that 240% average is 2.4x performance. People who make AMD slides are so clueless its insane.

Confusing as hell...

 

[Hail The Brain Slug]

Why does the 380 have halved L3 cache and only 16 Graphics cores? I am unaware of AMD disabling any L3 cache on a chiplet since Zen 3 debuted in 2020 and disabling 60% of the iGPU on the IOD seems unusual. Is the 380 even based on Strix Halo? Some kind of different part? I can't think of a way the Strix Point die could be configured to give 6 cores + 16MB L3, so I am even more confused at what it could be based on.

 

[Bigos]

And they said 1.4x faster which is 140% faster unless I'm bad at percentages. So it seems technically correct but unusual wording.

When you say something is 2x faster do you really think someone meant 200% faster or 3x as fast? I am not a native English speaker, but this makes no sense to me at all.

 

[GTracing]

View attachment 114349

AMD doesn't seem to understand that 240% average is 2.4x performance. People who make AMD slides are so clueless its insane.

Where is that slide from? The live-stream on YouTube says "260% average faster rendering on that chart. Timestamped link:

 

[adroc_thurston]

Is the 380 even based on Strix Halo

Yeah it's basically the poverty entry to the lands of Big Laptop Bandwidth and mem capacity.

 

[desrever]

Why does the 380 have halved L3 cache and only 16 Graphics cores?

Probably disables the MAL cache for the 380 cause it doesn't need it with so few GPU cores.

 

[Hail The Brain Slug]

Probably disables the MAL cache for the 380 cause it doesn't need it with so few GPU cores.

That's the cache on the Zen 5 CCD that's halved, not on the IOD. I don't think they've ever done that. I don't even know how they could disable half the L3 with 6 cores, maybe they could disable the slices adjacent to the 2 disabled cores but that would leave them with 24MB L3.

Does someone know more about this?

 

[gdansk]

When you say something is 2x faster do you really think someone meant 200% faster or 3x as fast? I am not a native English speaker, but this makes no sense to me at all.

I would assume faster was used for a reason. If they meant it was 1.4x the performance I would expect them to say that rather than faster.

But I would also expect a native English speaker to use 140% faster instead of 1.4x faster...

 

[desrever]

That's the cache on the Zen 5 CCD that's halved, not on the IOD.

Thats weird then. I guess they disabled half the L3 or something.

 

[Hail The Brain Slug]

Thats weird then. I guess they disabled half the L3 or something.

That is what I am saying I am unaware of them ever doing and I don't know if that's even possible. Someone's gotta know more about this

 

[soresu]

Can't quickly find the info on Zen5 X3D, does it have the v cache die on the bottom or the top of the stack in Granite Ridge?

 

[Hail The Brain Slug]

Can't quickly find the info on Zen5 X3D, does it have the v cache die on the bottom or the top of the stack in Granite Ridge?

V$ is underneath.

 

[Hitman928]

And they said 1.4x faster which is 140% faster unless I'm bad at percentages. So it seems technically correct but unusual wording.

No, 1.4x faster is the same as 40% faster.

What you could say is the product has 140% the performance compared to the competitor at 100%. This would be the same as 40% faster.

But in AMD’s case, they are at 240% compared to LNL at 100%, so they are 2.4x faster or 140% faster.

 

[Philste]

So Krackan indeed looks like a joke, especially the Ryzen 5 SKU. That thing is more of a Mendecino successor than anything.