AMD Readies Even More Derivatives of the 4 nm "Phoenix" Processor Silicon

[btarunr]

AMD's "Phoenix" monolithic processor silicon drives the company's Ryzen 7040 series mobile processor lineup, and possible some of its upcoming Ryzen 7000G desktop processor models. It is the first chip from the AMD camp to feature an AI accelerator, besides up to 8 "Zen 4" CPU cores, and a large iGPU based on the latest RDNA3 graphics architecture, with up to 12 compute units, the latest display I/O and media acceleration capabilities. Over the course of its lifecycle, AMD realized that it can't use the nearly 200 mm² silicon built on the expensive 4 nm node to power lower-end processor SKUs, and so developed the smaller 137 mm² "Phoenix 2" silicon that lacks the AI accelerator, has a smaller iGPU with just 4 compute units, and a unique hybrid CPU with 2 "Zen 4" and 4 "Zen 4c" cores. We're now hearing that the company is designing even more derivatives.

The PCI ID Repository discovered two new IDs believed to reference the iGPU models of "Phoenix 3" and "Phoenix 4" chips. At this point we have no clue what the two chips could be, and what the mixture of their CPU, iGPU, and AI accelerator components could be, especially given that AMD is able to carve out Ryzen 3 SKUs from "Phoenix 2." We speculate that "Phoenix 3" and "Phoenix 4" could reference rebranding such as "Escher," although it could even be entirely new chips with different combinations of "Zen 4" and "Zen 4c" cores.



View at TechPowerUp Main Site | Source

 

[ratirt]

I will need new laptop at some point I would hope for the Phoenix to be the one to go with. I need something to play a casually. Fingers crossed for price to be "OK".

 

[Space Lynx]

x3d APU 4nm please. imagine that in Deck 2, or 2nm x3d APU in Deck 2... oh my what a dream.

 

[ratirt]

x3d APU 4nm please. imagine that in Deck 2, or 2nm x3d APU in Deck 2... oh my what a dream.

Is there a plan for the x3d's in laptops? I mean it does make sense in terms of lower clocks and lower power consumption etc. but is the extra cache really beneficial for an APU in games? I think not but I could be wrong about it. Faster mem is always welcomed but I think it will not have a huge impact on the graphics chip performance.

 

[Space Lynx]

Is there a plan for the x3d's in laptops? I mean it does make sense in terms of lower clocks and lower power consumption etc. but is the extra cache really beneficial for an APU in games? I think not but I could be wrong about it. Faster mem is always welcomed but I think it will not have a huge impact on the graphics chip performance.

I honestly have no idea. I hope they are testing it out in the RnD section of AMD though.

 

[ratirt]

I could be really excited about the APU's and how much of an impact the x3d would have on the graphics since being excited about the discrete graphics cards is foolish nowadays. Also, I'd retire my old laptop for something with more oomph.

 

[tabascosauz]

I honestly have no idea. I hope they are testing it out in the RnD section of AMD though.

The engineering needed for this would be massive and pointless. Vcache stack is designed to sit on (or slightly spilling over, for Zen 4 CCD version) a 32MB L3 - APUs have only ever had half that space. Phoenix is also almost 250% the size of a Zen 4 CCD, the silicon spacer needs to be many times larger to cover Phoenix and make it contact a cooler/IHS properly, and negatively affects thermals for all of that die space. iGPU and APU SOC traditionally runs relatively cool in APU compared to core, but 780M is getting up there in terms of power consumption.

Not to mention that Vcache is currently still a N7 piece of silicon, although it was redesigned to fit on the N5 Zen 4 CCD. Phoenix is N4.

They're not going to do something as stupid as torpedoing Raphael/Dragon Range 8-core's entire reason for existing, by engineering Vcache into Phoenix.

 

[leezhiran]

Is there a plan for the x3d's in laptops? I mean it does make sense in terms of lower clocks and lower power consumption etc. but is the extra cache really beneficial for an APU in games? I think not but I could be wrong about it. Faster mem is always welcomed but I think it will not have a huge impact on the graphics chip performance.

AMD already launched 7945hx3d.

 

[wolf]

Something with the 780M, minimum 6C/12T and with 32GB of LP DDR5 7500+ would be very nice. I really think they need a 32GB version of the ROG Ally and Legion Go, 16GB is fine for now, but these things have potential to last a very long time and have great general PC use chops, so if it's going to be fast LP DDR5, I want a more comfortable amount of it.

 

[Firedrops]

at this rate, it wouldn't be long before AMD has more product SKUs on paper than actual physically manufactured units.

AMD already launched 7945hx3d.

"""""already launched"""""

 

[TumbleGeorge]

I could be really excited about the APU's and how much of an impact the x3d would have on the graphics since being excited about the discrete graphics cards is foolish nowadays. Also, I'd retire my old laptop for something with more oomph.

Higher price in the range of modern cheapest dGPU because x3d?

 

[T1beriu]

x3d APU 4nm please. imagine that in Deck 2, or 2nm x3d APU in Deck 2... oh my what a dream.

X3D doesn't make any sense for low power APUs. This tech makes sense when the GPU is bottlenecked by the CPU.
In this kind of product the power and GPU memory bandwidth is the bottleneck.

 

[Space Lynx]

X3D doesn't make any sense for low power APUs. This tech makes sense when the GPU is bottlenecked by the CPU.
In this kind of product the power and GPU memory bandwidth is the bottleneck.

Linux (SteamOS) handles memory bandwidth differently though I thought? @Easy Rhino am I mistaken or doesn't Steam only have like 1gb of vram cause LInux does vram differently at the kernel level?

 

[ratirt]

X3D doesn't make any sense for low power APUs. This tech makes sense when the GPU is bottlenecked by the CPU.
In this kind of product the power and GPU memory bandwidth is the bottleneck.

I wouldn't be so sure. And your statement that extra cache helps when GPU is bottle-necked by CPU is also not quite right. That would depend on the game if the game benefits from the extra cache. Not all scenarios will have that boost when more vcache is present.
The cache mem is faster than any memory so in theory it should help igpu to speed things up. I'm not sure by how much.

 

[TumbleGeorge]

As far as RAM throughput is concerned. Data to L3 coming from RAM is not infinitely fast.

 

[Haile Selassie]

Inventory piling up, eh, dear AMD?

 

[ToTTenTranz]

For what it's worth, @Kepler_L2 on X/Twitter mentioned Phoenix3 a couple of weeks ago as it being a 9W Van Gogh successor (AFAIK not to mistake with Van Gogh's N6 shrink called Sephiroth).

https://twitter.com/i/web/status/1720169521653166297

https://twitter.com/i/web/status/1720236069470498872

https://twitter.com/i/web/status/1720387524944605481

Inventory piling up, eh, dear AMD?

Nonsense. AMD is selling all Phoenix chips they can make and there's good proof of that.

https://twitter.com/i/web/status/1713845393795264823

 

[Daven]

Inventory piling up, eh, dear AMD?

AMD is selling everything they can make and are now turning to those parts with defects (limited power/clocks) and disabled silicon that typically pile up enough to launch new products from these parts. Enough of these subpar parts are available around a year after the initial launch.

Nvidia is doing this with its Super series. Intel does this by having dozens of SKUs.

AMDs problem is a chicken and egg problem. You need money to buy fab capacity but you need fab capacity to make more chips and therefore make more money. They buy all the chips they can afford, sell EVERY one of them and hope their margins are high enough to afford more chips in the next production round. This is the essence of AMDs chiplet strategy to make more chips for the least amount of money.

 

[TheinsanegamerN]

X3D doesn't make any sense for low power APUs. This tech makes sense when the GPU is bottlenecked by the CPU.
In this kind of product the power and GPU memory bandwidth is the bottleneck.

It makes PERFECT sense. The 945xd saw major iGPU performance improvements, and that is a lowly 610M. The 780M would benefit even more.

And the larger the cache, the less hits you need to DRAM, increasing efficiency. This is computing 101.

 

[TumbleGeorge]

About gains from x3d for iGPU. Hmm, not as expected here.

 

[ratirt]

About gains from x3d for iGPU. Hmm, not as expected here.

This is only 2 graphics cores. I'd rather have an APU test than a desktop CPU with a basic GPU but apparently not that much difference. Wonder if 12 graphics units will make a difference.

 

[Wirko]

About gains from x3d for iGPU. Hmm, not as expected here.

The iGPU in desktop Ryzens has its own cache and can not access the CPU cores' L3. The same is true of AMD's monolithic APUs. As is nicely explained in the article you've linked to,

However, the iGPU resides on the I/O die and is a self-contained unit with its own L2 cache. As such, it doesn't access the L3 cache on the compute die in any meaningful way (diagram further below in the article). Therefore, any performance gains couldn't come from additional cache bandwidth.

 

[TumbleGeorge]

The iGPU in desktop Ryzens has its own cache and can not access the CPU cores' L3. The same is true of AMD's monolithic APUs. As is nicely explained in the article you've linked to,

So, is it even possible and cost-effective, and especially how to connect the cache to the iGPU in another way?