performance on GLSL "for" instruction

Hi I'm doing fractal work using GLSL, and I need to use the "for" instruction. Is is better (performance-w开发者_如何学Cise) to do a for instruction n times, or to redirect the ouput of the shader on a FBO, and apply the shader n times on a FBO? Thanks :)

Applying a shader n times will definitively be more expensive. It will not only be more expensive due to the actual shader, but also due to texture fetch and ROP. Plus, rebinding buffers and synchronization.