Similar things have been done in the past.. but with very poor results.
GPU's are preferred for duties such as executing shader code, hardware transform/lighting and such due to their limited design. As they are specialized for such operations, they can do these things several hundred fold greater than a generalized processor like a CPU.
Several years back, a wrapper for Direct3D T&L was written to emulate this limited functionality using MMX/SSE (Intel) or 3DNow! (AMD).. and the results were CPU's could only muster like 1/20th the performance of the lowest end GPU's doing the same (i.e. GF2-class hardware). CPU's just can't grab/calculate high volumes of small matrices in/out of cache/main memory as quickly as this occurs on dedicated, small instruction GPU into faster videoram/video memory buses.
Microsoft ships a "reference rasterizer" if you're interested in furthering your knowledge on such things. Over time, this has been updated to take advantage of CPU instructions better suited for these things (i.e. SSE2 and such). If you're interested, you can freely download Microsoft's Direct3D SDK from msdn and it includes example Direct3D applications where you can switch between the "reference rasterizer" (i.e. uses the CPU totally for all 3D calculations) vs. any capable on-board 3d chip. It makes for some interesting experience.
You can also download the entire Visual Studio Express development system in your choice of languages (although I'd recommend C# or C++) along with the SDK and be writing your own Direct3D games as well as pour through the dozens of examples it ships with (along with full source code).. all for free.
Keep up your hunger for knowledge and outside the box thinking, though greatman05... and I'd encourage you, at your advanced level for a 14-year old, to pursue such avenues. It's always been outside-the-box thinking in years past that have brought us to what we have today- including GPU's, 3d graphics and even Unix.
