Comment by senkora

For PlayStation APUs, it's likely that AMD presents a menu of options and Sony chooses which components they want. For PS5 Pro, the CPU is unchanged from PS5 because Sony doesn't feel the need for anything faster. A newer CPU would take more area. But Sony really wanted better raytracing and AI so they chose RDNA 3.9 or whatever for the GPU. I suspect the cores are all mostly synthesized so they can support any process and Infinity Fabric is compatible enough that you can mix and match new and old cores.

> They (AMD/Sony) shoehorned the RDNA 3/3.5 GPU architecture onto an older Zen 2 core

The original core was already a custom configuration. I don't see why it seems odd that the new version would be a custom configuration based on the previous one.

> with a different process node

This doesn't apply to the PS5 SoC, but is general to AMD's methodology.

AMD has been using an off-chip interposer setup for multiple generations now. They did this specifically to allow for different process nodes for different chips.

It's cheaper (and there are more fab options) to produce chips at a lower process node. If there's no reason to update the CPU, it would make sense to keep it on the cheaper option.

In regards to the PS5 and Xbox SoCs, specifically.

The entirety of the SoC is fabbed at the same process node. A core designed for a 14nm process node and then fabbed at 7nm (assuming drastic changes weren't needed to make it function at the lower node) is going to be much smaller and run cooler on that node size. This is cheaper and leaves more space in the total footprint for the GPU-specific and auxiliary logic cores. Same rule applies above, why use more if it's not needed.

> they felt like making a frankenAPU

All of the game console chips are "frankenAPUs".

> Especially since the APUs are usually monolithic (vs chiplet) in design and share a memory controller.

"Monolithic" vs "chiplet" is an arbitrary distinction, in this case. The individual logic cores are still independent and joined together with interposers and glue logic. This is clear from the die shots:

https://videocardz.com/newz/sony-playstation-5-soc-die-pictu...

To return to the previous point, look at the space dedicated to the CCXs. The Zen2 has ~1.9bln transistors, the Zen3 ~4.1bln, the Zen4 ~6.6bln, etc. To use a newer core would double or triple that space. Increasing the total die size, making it more expensive per chip and increasing the defect rate.

> Surely it would have been easier/cheaper to put in 8 zen 4c/5c cores and call it a day.

Definitely not.

> I'm pretty sure I'm just missing something obvious...

Nothing about chip design is obvious.

PS5 Pro makes no sense, yes.

Most studios aren't even able to push current PS 5 to its limit, given current development schedules and budgets.

PS 5 Pro is for the same target audience as PS 4 Pro, hardcode console fans that will buy whatever the console vendor puts out, and Sony needs to improve their margins.

The comment above is elaborating on x86 micro-architecture, the differences between how the CPU handles x86 instructions specifically.

The overall system architecture is different between PC, which has discrete memory systems for the CPU and GPU, and a very long pathway between GPU memory and system/CPU memory, versus today's consoles which have unified memory for CPU+GPU, and optimized pathways for loading from persistent storage too.

Consolesuse their own graphics APIs, but you would have any vendor you contract with for graphics support your native graphics API and everything would be "fine". PS5 games use GNM/GNMX Playstation proprietary graphics APIs. Usually PC ports of console native games re-implement the rendering engine using the PC graphics APIs like DirectX or Vulkan. The re-implementation is probably less efficient and less tuned.

Horizon Forbidden West was ported from PS to PC. Decima is an engine from Sony’s first-party studio, so it's understandable that their development process would lean more towards PS's internal architecture rather than the more common GPUs on the market. Of course, even general-purpose engines can perform better on PS5, AMD, or NV. But, for these engines, they have less information about how customers will use the engines, so there's less infomation can be used to optimize. On the other side, customers using these engines often don’t have enough experience optimizing sufficiently for each platform. None of this is absolute, but I think this logic is reasonable.

For game developers using these engines, if they take optimization seriously, they typically make adjustments to lighting, LOD, loading, and model details or shaders on console platforms to achieve a similar visual effect while meeting the targeted performance goals. This is why you usually get better performance on a console at the same price point compared to a PC, aside from the subsidies provided by Sony.

> Horizon Forbidden West needing a much better GPU on PC to run at the same level of fidelity as the PS5.

not being expected to run with variable refresh rate\interleaving and accepting 30\60 fps in best case situations?

> They almost all use off-the-shelf middleware and game engines that are built to support all of the platforms. The companies that do go through that effort tend to have very notable releases.

There are still manual optimizations that can be done (non-temporal writes where appropriate for example), but nothing like the painstaking removal of Load-Hit-Stores and cache control of the 360/PS3 era.