fpbuffer question

I am trying to use the GPU for scientific computing and was looking at an fpbuffer as a good rendering target to use before reading the results back to the GPU. I am finding it difficult to get any good information on what the fpbuffer offers in terms of functionality. The vectors that I want to work on are going to need more than 8 bit precision per component, can I use a C float (x86 win) level of precision per component? i.e. do fpbuffers offer higher levels of precision than a regular pbuffer?

thanks for any help on this.

On NVIDIA hardware (NV30+) you can create a floating point pbuffer (using the GL_NV_float_buffer extension) where each component is 16 or 32 bits.


Although, the GPU does not have the same control over rounding modes so you have to be careful about the real precision of your results. It is not exactly the same as IEEE 754.