The Cell processor is a step towards creating a hybrid CPU/GPU; it is very good at computing loads of floating point operations very fast - and does away with a graphics card.
However, most gaming logics require short, branched operations that such a processor doesn't handle well.
As such, the very ancient idea of a core CPU targeted at computing integers with a FP coprocessor is coming back today as graphic cards with programmable shaders (which aim at computing large amounts of FP data as fast as possible).
'Current' processors have an integrated FPU; however, it is much less powerful than a graphic card's dedicated GPU, making software rendering of a scene much slower on those. It doesn't mean processors can't render a scene, it just means they need MUCH more time.
Note: softwares like 3D studio used to be CPU-only. Even now a 3D card is used for pre-rendering, not for final rendering. This is due to the various optimizations used in the card and in the driver that make several computations approximative and a bit unreliable.
Professional graphic cards in fact come with special drivers that make such optimizations inactive and stuff like OpenGL texture clamping more precise - so as to do away with software rendering as often as possible.