I must be missing something....
I don't know much about Fusion or Torrenza beyond the newsbites I duly skim like everyone else, but how is any of this news? Fusion is a CPU plus some GPU "accelerators" (aka stream processors) onboard. If AMD is going to push Torrenza, they may as well use some of their own tech to kickstart it (alongside Clearstream's offerings), and what better (read: already developed and cheap to reconfigure) part than an offshoot of the "accelerator" part of Fusion?
So, Fusion remains a CPU+GPU (GPU = stream processors + ROP), and anything you plug into a Torrenza-compatible socket would be just stream processors that have access to their own slice of system RAM (essentially, cheap FireGL solutions). Would you be able to use these seeming GPPUs (not GPGPUs, b/c I'm thinking they won't have ROPs) to boost Fusion's integrated GPU's shader power? Seems overly complicated compared to just buying a faster discrete GPU, but then that's why they're spending money on Crossfire development, I guess.
How this differs from HTX, I don't know. Again, maybe it's simply a cheaper option, or a more flexible one in that you can use both sockets for either dual CPUs or one CPU and an accelerator. I'd be surprised if AMD meant for you to have and use simultaneously an IGP on both the NB and on the CPU (Fusion), or even two Fusions in one system. Fusion still seems like an attractive budget solution (no discrete NB or second socket), with the performance solution being the dual-socket MB pairing a quadcore CPU with a GPPU ("accelerator" or stream processor) in the other socket and probably a discrete NB + GPU handling the likely trivial rendering tasks intended for a config like that.
But maybe I'm missing a possibility.
