Just noticed that Ati's drivers support http://www.opengl.org/registry/specs/ARB/half_float_vertex.txt
16-Bit floating point makes sense for computer graphics, 32-Bit (or higher) precision is quite an overkill for Normals or Texture Coordinates. Do you think it would make sense to add new types to OpenTK like this?
They don't necessarily have to support any arithmetic operations (use float|double for that), I'm rather thinking about a container that does the pack/unpack and can be easily added to Vertex structs.
Edit: If I'm understanding http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/... page 21, section 1.5 correctly, we could use
short to store it. SSE5 will support float->half and half->float conversion, so future iterations of Mono could improve conversion speed aswell.