There is a very cool book :
FAB: The Coming Revolution on Your Desktop--From Personal Computers to Personal Fabrication (by Neil Gershenfeld)
that describes the use of computers to personal manufacture things.
3D in non-gaming world? I guess there is some sort of convergence of technologies between gaming and serious computations. Pretty much 3D mathematics is used in physics computations of pretty much anything: dynamics, strength of structures, computational fluid dynamics, magnetic fields, CAD, CAM and meshing as a common denominator of all these things etc. There is an enormous need to calculate things and I can see how next generations of video cards may be helpful. There is concept outhere, callded general purpose calculations GPU (GPGPU), which means unloading the calculations from CPU to GPU so they can be done faster because of the parallel execution that video cards inherently do.
I would like to see video cards in the future that have pixel shaders capable of double precision and easier programming infrastructure. I'm not very familiar but shouldn't the unified shaders take care of this problem?
Anyway, currently there are structural analysis and comutational fluid dynamics software that are written for parrallel processing machines/clusters and they can run in single precision and give pretty good results for certain configurations if a good quality mesh of the simulation is provided. They run for hours on multiple CPU machines/clusters. In the future they may use the GPU's to do that. All these computational software end up in solving iteratively massive (millions or billions of unknowns) sparse linear or nonlinear systems of equations. Parallel execution is the way to go, and video cards processing of 3d geometry data or pixels is inherently parallel. And you can relatively easy map the physical data to 3d vertex or pixel data, then let the multiple pipelines vertex and pixel custom written shaders that implement the physics calculations solve it.