OSDev.org

I know that anybody has seen that there are now many graphical environments in general that take up too much resources, yet there are other, like the newer 3D games that take less of them and even can work without using specialized video cards.

So, would it be worthwhile to try to do a full 3D library that doesn't use ever 3D hardware capabilities? Wouldn't it be enough to use these new multicore 64-bit and 128-bit-value SIMD (single instruction, multiple data/results) instructions and characteristics?

If so, it would be so nice to start to think on doing something like Beryl (for those who'd like to see a bit more: http://www.youtube.com/watch?v=ZD7QraljRfM) but that runs even on the cheapest hardware.

Does anybody know some reason NOT to do such low-end codebase and still even so not being able to get nVidia worth GFX results?

You should be aware of the fact that there are many many many "cheats" that you can pull off to make things look nice and get away with it fast. A generic 3D library could most likely not cover all of those.

Still, while nvidia and ati keep their card's specs to themselves, having a free software 3D library capable of decent speed is most certainly not a bad thing to have.

AFAIK, Mesa3D does that already. It's a bit slower (quite a bit, tbh) than using a physical card with OpenGL but it works. You can try glxgears (google) for testing your implementation/installation on your OS.