Apple 16" MacBook Pro (M2 Max, Space Gray)

Embedded usually refers to having the DRAM on the same piece of silicon. (For eg the old Nintendo GameCube has the DRAM on chip). The M1 chip have the DRAM on 1 to 4 separate silicon dies but they are packaged together. That on-package interconnect is slower than embedded, but is still faster than off chip. The unified memory model refers to the logical memory model provided to the devices. In the case of a nvidia GPU, the intel CPU does not have access to all the GPU memory hence its not unified. What is referred to here is that 4 separate DRAM module on a M chip can be accessed coherently and consistently by different components like the cpu and GPU. Note that this is not always optimal because say GPU need to have high bandwidth and the CPU needs low latency - most system will segment the memory for GPU and CPU , piece of mem can be optimized for bandowidth and other for latency. Having a unified memory means having the lowest common factor. Thats is how the integrated graphic controllers from Intel use in previously Ng ration, it uses the system DRAM for both graphics and CPU.