It's a circle
<<when memory, fabric interconnect, coprocessors, and likely central processors will all be crunched down to single chip packages, we won't even be talking about coprocessors any more.>>
The history shows that there is a circle : we design some systems which plug in into the machine. Then, after some years, we are able to add these functionalities inside the CPU, so we move them there.
Wait another 10 years and again we have more powerful systems (than the one available in the CPU) on some additional board... everybody starts using them, we make transistors smaller or die bigger or whatever, and the CPU includes this functionality.
You want floating point ? Buy the card and plug it into you PDP-11, oh well, no, add a co-processor, it's in a chip now, oh wait, let's move the FPU into the CPU, okay, it's there, but only one FPU ? Need many ? Buy this external card, ah wait, AMD have some APU's now, not powerful enough ? Buy a card then, etc. etc.