But now let's totally eliminate the barrier between graphics, sound and all other processors. Instead of limited communications channels and local memory, have distributed shared memory (DSM) and totally free communication between everything.
This sounds a lot like NUMA. Which, I might add, absolutely requires differentiating between local and non-local memory, since the latter is much slower.
Thus, memory can open a connection to the GPU,
Like GPUs have done since the time of AGP? Or did you mean memory will simply send some random data for no particular reason?
the GPU can talk to the disk,
For what purpose? Do you plan to write a file system driver that runs on the GPU? To accomplish... what, exactly speaking?
Ethernet cards can write direct to buffers rather than going via software (RDMA and OpenSockets concepts, just generalized).
Haven't they done this a long time now? In fact, don't all devices that do significant IO use direct memory access?
What room, in such a design, for a CPU? Everything can be outsourced.
And the part that keeps track of the overall program execution state and issues these outsources tasks to other components is, for all intents and purposes, a CPU.
Have the router elements take care of heat and congestion issues, rather than compilers.
...What the heck are you talking about?
And this is marketspeak? Marketspeak for what? Name me a market that wants to eliminate complexity and abandon planned obsolescence in favour of a schizophrenic cross between a parallel Turing machine, a vector computer and a Beowulf cluster.
None does. It's the "schizophrenic" part that's the killer. Which is why, if you need to sell garbage anyway, you litter your product description with enough trendy buzzwords to convince technologically illiterate that it's cutting edge high tech. Which, if you are trying to polish a particularly smelly turd for a sale, can end up using almost all of them. And that can have great synergy with illusion-challenged human resources seeking a solution for cynicism management.