The thing is, GPGPU is not some theoretical dream for the future, it's already being used, and can only be used MORE in the future.
The basic programming techniques are not unique to PS4 here, it is not like Cell, but something that will be widely used, including on Xbone.
More capacity for GPGPU like PS4 has (on par with the highest end PC cards in this sphere) lets you do more things with it,
but XBone has mostly the same underlying GPGPU technology, just 1/4 the capacity for # of GPGPU threads (program functions).
The unique thing about PS4 GPGPU, besides it's high amount of GPGPU capability is it's memory system architecture,
which will let all the GPGPU threads easily share data with threads on the main CPU, using the best of each side's capabilities.
That will increase the number of things it will be useful for, allowing it to easily influence game functions that reside on the CPU.
Here is some links that came up from Google on what you can do with GPGPU:
http://www.slideshare.net/zlatan4177/gpgpu-algorithms-in-games
particles, fluids, destruction (physics), graphics post processing,
ambient occlusion, tiled lighting,
pathfinding "can saturate GPU without impacting framerate"
http://www.tomshardware.com/reviews/directcompute-opencl-gpu-acceleration,3146-2.html
Example of GPGPU Ambient Occlusion
http://mesh.brown.edu/taubin/pdfs/Zhao-ivc2011.pdf
these guys use GPGPU to allow computer vision to "see in 3d" from stereo video feeds,
so I might guess that GPGPU might be applicable to the reverse, creating stereo video
feeds for a 3-d experience, i.e. Sony's Morpheus headset. Normally that's done by rendering
stereo video feeds more or less separately, but since the scene images are mostly identical,
there seems scope to have some function render both video channels with less work than independent rendering.
http://blogs.msdn.com/b/shawnhar/archive/2008/12/31/pixel-perfect-collision-detection-using-gpu-occlusion-queries.aspx
MS themself discuss how GPGPU can do collision detection
https://developer.nvidia.com/gpu-ai-path-finding?display[%24ne]=
NVIDIA on GPGPU Pathfinding AI
http://www.nvidia.com/content/GTC/posters/20_Kider_High_Dimensional_Planning_on_the_GPU.pdf
Dense, but seems applicable to realistic physics simulation of skeletal systems, etc. (complex multidimensional movement)
