I'm looking to implement a set of classes to handling loading, animating, and drawing skinned meshes. I understand the theory behind the principle: vertices are mapped to one or more bones with weights ranging up to 100%; The bones are arranged in a hierarchy so that bones further down in the structure are affected by all preceding ones; the hierarchy can be implemented using a matrix stack; and so on... What I'm wondering is how the animating and drawing process should be implemented. What I mean is, I've been reading Jim Adams' book "Advanced Animation with DirectX" where he makes a copy of the mesh each frame and updates the vertices on the CPU. This seems like a slow way of doing things, since I would assume much of the work could be offloaded to the GPU. Is there an efficient way of applying the bone transformation matrices and rendering all in one step in either the fixed-function pipeline or a shader?