Ok, I bound a seperate buffer for the output vertices. I did get an error INVALID_OPERATION for the first glDrawArrays. Now everything is processed. But can't you define a single VBO as an input and output buffer for transform feedback. Or do you just copy the the feedback buffer into the VBO for drawing?
I mean, isn't this transform feedback could be used for? Updating VBOs just on the GPU through shaders?
For example, if you wanted to make a particle system that is just updated on the GPU with millions of paticles; copying all that data for every transformation is very expensive. Why not just write it into the same same VBO then?