The reason that comes to my mind is to save memory to transfer from the CPU to the GPU. If you do your tesselation inside the vertex shader, then you can send

The reason that comes to my mind is to save memory to transfer from the CPU to the GPU. If you do your tesselation inside the vertex shader, then you can send a low-poly mesh to the graphics card, which saves a lot on vertex buffers (e.g. uv coordinates and other per-vertex attributes). The vertex shader still emits the same number of vertices to the rest of the pipeline, but inter-stage bandwidth is more plentiful than CPU-GPU bandwidth so I can see that coming out ahead. I’m not an expert though. Perhaps someone with a better understanding can clear this up, I’m curious too…