I seem to have landed on OpenTK. I think it gives more or less direct access to the OpenGL API and doesn't require loads of dependencies.
It is comparatively easy to understand. It requires few (and understandable) lines of code to get started. It doesn't hold the objects for me so I'm free to change anything for each pass, which is great because I'm mainly working with unsafe pointers to memory.
It is of course difficult to combine speed with ease of use. Speed requires talking directly to the 3D API while ease of use requres abstraction. Therefore this must be considered a lower level API than some of the others I've tried. If I wanted to do some prefab character animation then XNA would probably be a better choice, but for my use (described above) this seems very promising so far (4-5 hours of hacking).
Some sample code:
private void Render()
{
// Every frame
GL.MatrixMode(MatrixMode.Modelview);
GL.Clear(ClearBufferMask.ColorBufferBit | ClearBufferMask.DepthBufferBit);
GL.LoadMatrix(ref cameraMatrix);
GL.Begin(BeginMode.Points);
// Here I can add lots of points. I add 200k without any big problem.
// It seems these points could have been passed in as an array pointer too,
// but I'll look at that later.
GL.Vertex3(x2, y2, z2);
GL.End();
glControl.SwapBuffers();
}
