Apple's claim isnt 100% untrue tbh. It depends on the operations actually. Arm processors have at least 12 registers to contain data or references to memory. A program does need more ram space on a x86 processor, as it only has like 4 registers afaik ( correct me if im wrong! ) so it needs to push data more often to the stack.

This means that the m* processors has to generally save less info in memory. However, data is still data and you still need memory to contain the data youre processing so you still need the ram. So like, when doing video work that apple claim is utter bullshit. Raw calculations however might not be so much bullshit

The SoC and memory are separate dies with different manufacturing processes. In the case of M2 it was TSMC for the SoC and SK Hynix for the memory.

When it comes time to package them together, the SoC and memory are soldered to a interposer layer. So the only difference is which size memory chips they solder together for the different memory configurations available.

That's like building a fast car that can only go straight. It's impressive but short-sighted and therefore stupid.

Apple uses a unified memory where the memory chips are embedded on the SoC in the first place. The memory modules are on the same silicon wafer the chip is cut from, not separately on the Mobo

This is 100% false. All Apple Silicon Macs use standard LPDDR4X or LPDDR5 memory chips, the same as are used in other computers, which are soldered on a PCB next to the SoC. They are not on the same die. The high memory bandwidth on M1/M2/M3 comes from having a lot of memory controllers built into the SoC -- it's akin to a PC with an 8+ channel memory setup. As far as I'm aware, there's nothing technically preventing Apple from making an Apple Silicon mac with socketed memory again, other than those sweet sweet profits for shareholders.