It is, due to latency and bandwidth. But, depending on the PCIe lanes and generation, it can fetch a decent amount of data.
You can watch these two videos for when there is a video memory deficit that pushes data into system memory. Using a PCIe 5.0 mobo isnt going to have any
In this video its clearly demonstrated, newer PCIe but lower memory bandwidth wont improve anything.
A 16GB 5060ti using PCIe 3.0 outperforms an 8GB 5060 on PCIe 5.0 by over 60% since no data has to be fetched from system memory, every necessary asset in resident in vram. If the Steam Machine hits vram deficit and places that data in system memory it wont be able to match a system with a better gpu more memory and memory bandwidth despite pcie 5.0.
This video just demonstrates fundamentally why a vram memory deficit is bad. even on the same pcie board, fetching data into system memory during a vram memory deficit yields no noticeable benefit in performance.
So the Steam Machine just physically has a narrower bus width, lower capacity and not even PCIe 5.0 can compensate for that, base PS5 alone will always outperform it despite PCIe 4.0 since no gaming assets are ever resident in low bandwidth memory. If it had more VRAM then there would be significant gains in terms of texture quality at least. Like say 10-12GB of gddr6, but there's about 4.5GB more high bandwidth memory in the base PS5.
Contention of memory bandwidth is a problem is having one single pool, accesses by the CPU and the GPU. It's just a fact that even Sony has addressed in their presentations of consoles.
PC has some duplication of data, between pools, but it's not that big of a deal. The real problem with having 2 pools of memory is going through the PCIe bus.
You're comparing a first order constraint(GPU bandwidth, VRAM size) to a second order theoretical challenge from early PS4 days. Memory contention between the CPU and GPU if unaddressed by the PS5 would still outperform the Steam Machine's first order contraints of lower VRAM and bandwidth. But Sony has actually addressed any potential theoretical issues by designing the memory controller to smartly schedule access such that the CPU and GPU never thrash each other. The PS5 unified memory has no memory contention issues. As well the software APIs have CPU/GPU scheduling so devs simply have to drop data into unified memory without any worry about such a second order constraint. Memory contention only came about during early PS4 days and even then it was about designing around a fixed budget than unified memory is fundamentally flawed.
Yes, 32Mb of L3 cache helps a lot with reducing memory accesses. Around 30% at 4K, Around 45% at 1440p and around 55% at 1080p.
This is AMD's own data. And was also corroborated by nvidia, when they started using a big L2 cache as well.
So in practical terms the APU in the Steam Machine will have as much bandwidth as the PS5. And it will have the advantage of lower latency for anything that hits the L3 on the GPU.
As you can see in the slide you attached, its estimated you'd need above 64MB of L3 cache to start seeing compelling bandwidth gains with 120MB being the sweet spot. 32MB doesnt hit that threshold. So its quite a stretch to say the Steam Machine will have as much memory bandwidth as the PS5.
The CPU on the Steam machine is a Zen4 CPU. And this has huge advantages to the Zen2 CPU on the PS5.
One is IPC, due to 2 generations of improvements. The other is that Zen4 is a monolithic CPU. While the Zen2 on the PS5 is a 4+4 core CPU, so when data has to be shared across the 2 core complexes, there is a big performance hit.
The Zen4 CPU on the Steam deck probably has the full 32MB of L3 cache. While the PS5 only has 4Mb+4MB.
Worst yet, the PS5 CPU has to deal with high latency GDDR6, that hovers at around 140ns. The Zen4 CPU will deal with DDR5 with a latency of around 70ns.
So the CPU on the PS5 not only will have a lot more cache misses, due to having much less cache and having an older CPU front-end, but when it has to go to system memory it takes a bigger performance hit due to high latency.
The other problem is that the CPU on the PS5 runs with a max clock speed of 3.5Ghz. While the CPU on the Steam machine has a boost of up to 4.8Ghz. That is another 37% deficit, on top of the IPC.
I agree the Zen 4 cores are better and the CPU in the Steam Machine will be better but I think in games like GTA 6, PS5 CPU will provide the edge with the 2.5 extra cores considering the game engine is built for highly parallel multithreaded workloads. So yes in the vast majority of games Steam machine CPU is better. I dont want to go into too many cpu related things for example only 2 of the 6 cores in the steam machine can boost up to 4.8Ghz, but yes steam machine has better cpu.