The fastest way to render the content generated in CUDA by OpenGL?

Accelerated Computing CUDA CUDA Programming and Performance
1
  1. Using a PBO and an OpenGL texture (old style);
  2. Using a OpenGL texture as a CUDA surface and rendering on a fullScreen quad (new style);
  3. Using a renderbuffer as a CUDA surface and rendering using glBlitFramebuffer.
    which one is fastest and the most effcient one? or any other better method?
    Thanks!