Question Zen 6 Speculation Thread

[luro]

It's N3-ish stuff versus the N7 VGH.

Oh, I thought it was on 18A for some reason

 

[adroc_thurston]

Oh, I thought it was on 18A for some reason

18A is very much an N3 class node.
A hair slower/less efficient than N3e.

 

[CouncilorIrissa]

18A is very much an N3 class node.
A hair slower/less efficient than N3e.

The B370/B390 die is straight up manufactured on N3E.

 

[inquiss]

The B370/B390 die is straight up manufactured on N3E.

No

 

[CouncilorIrissa]

No

Yes.

>built on external
N3E in this case.

 

[ToTTenTranz]

You do understand that the thingy there is like 2 nodes ahead?
I'm not even talking 50mm^2 of GFX area alone.

The Steam Deck still outperforms Strix and Phoenix / Hawk at 10W.

Point being Intel's iGPU scales down for lower budgets just fine.

 

[adroc_thurston]

No

Yeah.

The Steam Deck still outperforms Strix and Phoenix / Hawk at 10W.

Point being Intel's iGPU scales down for lower budgets just fine.

It doesn't.
Again, 2 shrinks, a ton more area and on average like 30% faster.
Embarassing.

 

[inquiss]

Yes.
View attachment 141331
>built on external
N3E in this case.

I stand corrected

 

[Saylick]

I stand corrected

FWIW, the weaker iGPU SKUs use Intel nodes:

 

[OneEng2]

I don't write assembly anymore, intrinsics made that a thing of the past, but I do look at the disassembly of my code. You can basically write assembly level code with tight c/c++.

I once wrote a small sample in asm, then C and ran them against each other. The C was FASTER than the ASM. When I disassembled it, I found that the C compiler used the registers better than I did. I quit writing ASM for anything other than inline C functions for very very specific usage after that 🙂.

9% is not that much. But do you know if Zen 6 is better than 9%?

Let's say 10% conservatively. On top of that, I hear the latest leaks are that clock speeds will top out at ~6.4Ghz (12%). I think the bigger effect (and likely not measured well by SPEC) will be the improved memory controller, faster memory, and larger L3 cache impacts that aren't necessarily "IPC"

In fact, I am wondering how important a "pure" IPC measurement is (say the algorithm fits into L1 to measure the MAX throughput possible).

So I am guessing that in most desktop applications we will see a 20-30% improvement over Zen 5.

I think in DC applications, we will see the 70% AMD is claiming.... and this is really a bigger deal than desktop anyway..... at least for AMD and their profit and revenue are concerned.

 

[adroc_thurston]

will be the improved memory controller, faster memory, and larger L3 cache impacts that aren't necessarily "IPC"

Those are hardly relevant outside of nT loads.
1t is IPC * clocks.

 

[Joe NYC]

Great now we are at a point where people are saying new ISA extensions can be delivered via microcode updates.

DLC

 

[511]

DLC

CPU Micro transaction?

 

[Kepler_L2]

CPU Micro transaction?

Intel Finally Reveals its Software Defined Silicon as Intel On Demand

Back in September 2021, reports about Intel working on something called SDSi or software defined silicon, started to appear. Now, over a year later, the company has finally launched its SDSi products under the Intel On Demand branding. Back then, we speculated about what features Intel would put...

www.techpowerup.com

 

[511]

Intel Finally Reveals its Software Defined Silicon as Intel On Demand

Back in September 2021, reports about Intel working on something called SDSi or software defined silicon, started to appear. Now, over a year later, the company has finally launched its SDSi products under the Intel On Demand branding. Back then, we speculated about what features Intel would put...

www.techpowerup.com

lmao

 

[Win2012R2]

These two sentences are mutually exclusive. If Valve is cheap, they're not going with Nvidia.

Intel will get Nvidia iGPUs soon - Nvidia will certainly be open to get into x86 handheld console market, they did it cut price for Nintendo who must be the cheapest company on this planet.

 

[ToTTenTranz]

Yeah.

It doesn't.
Again, 2 shrinks, a ton more area and on average like 30% faster.
Embarassing.

At 10W in gaming workloads, Hawk and Strix Point are also embarassing compared to Van Gogh considering their node generations as well. That doesn't mean RDNA3's gfx IP is terrible.


At 20-30W, Intel's Xe3 is surpassing RDNA 3/3.5 by 50 to 80%, or more when raytracing is involved and/or a XeSS is used for a lower base render with better IQ than FSR3.

 

[adroc_thurston]

At 10W in gaming workloads, Hawk and Strix Point are also embarassing compared to Van Gogh considering their node generations as well

Those weren't made for that.
Alas, LNL and PTL both were.

At 20-30W, Intel's Xe3 is surpassing RDNA 3/3.5 by 50 to 80%

A lot less, but that's hardly relevant.
AMD's not giving you anything btw.

 

[soresu]

Intel will get Nvidia iGPUs soon - Nvidia will certainly be open to get into x86 handheld console market, they did it cut price for Nintendo who must be the cheapest company on this planet.

Cut price on a 2 yr old SoC that was already outdated by the time it was announced the first time, let alone by the time it was announced for Switch and then released.

(it was effectively making money back on unused stock like with the AMD Steam Deck SoC deal)

The SoC for Switch 2 is even further out of date on the CPU side and was already 2 gens out of date on the GPU side - not to mention it's sporting a GPU cluster count inferior to most flagship phones, so it's not like they are stretching the budget from any side.

 

[ToTTenTranz]

Those weren't made for that.
Alas, LNL and PTL both were.

None of them, AMD nor Intel, were designed for gaming workloads at 10W.

Van Gogh OTOH does seem to be much more gaming-oriented, or at least much better balanced for gaming (only 4 CPU cores with big 8CU GPU, large GPU L2 per-CU, large RAM bandwidth per-CU, etc.).

A lot less, but that's hardly relevant.

It's very relevant considering the fact that, like you keep saying, AMD doesn't care.

 

[adroc_thurston]

None of them, AMD nor Intel, were designed for gaming workloads at 10W.

LNL was!

Van Gogh OTOH does seem to be much more gaming-oriented, or at least much better balanced for gaming (only 4 CPU cores with big 8CU GPU, large GPU L2 per-CU, large RAM bandwidth per-CU, etc.).

No, it was made for a meme Win10X tablet.

It's very relevant considering the fact that, like you keep saying, AMD doesn't care.

?
Dayum, brother. Just prep the cash for Olympic Ridge.