Earlier quoted context omitted.
3%? M2 = 100GB/sec M2 pro = 200GB/sec M2 max = 400GB/sec M1 Extreme = 800GB/sec. Most x86-64 laptops and desktops are in the 60-80GB/sec range. Some desktop or server only chip (threadripper, epyc, Xeons, a few intel workstation chips) have 150 or 300GB/sec. 400GB/sec vs 80GB/sec is quite a bit more than 3%, and is available in the mbp 14" or mbp 16".
That’s because Apple Silicon has a stupid-wide memory bus. It has little/nothing to do with the characteristics of the memory chips shipped in the box. They just put more modules in parallel than on x86. Apple could ship cheaper memory modules with the same stupid-wide bus and achieve nearly the same performance for a fraction of the cost. And that’s what I want. Give me 4x as much DDR4 on the same 1024 bit wide bus,…
Sure it's easy to ask for 1024 bit wide memory bus, but it's not cheap. Done with the traditional x86-64 method you'll need two 280 watt CPUs (a few $k each), on a $1k motherboard, with 24 dimms, a 1kw power supply, ear plugs, and at least $10k. The minimum physical size is going to be substantial since you need two large CPU sockets (6096 pins each), 24 dimm sockets, and of course a substantial number of fans, dimms, and power supply. Even managing airflow over 800 watts or so of chips isn't going to be easy, cheap, or quiet.
Because of the above size, cost, noise, and heat it's targeted at enterprise and cloud folks with the related cost, service contract, and sales pipelines.
Or you could buy an Apple studio and hide it somewhere, you'll likely not be able to find it by listening for it.