GPU Architecture

Hi, im a research student with sydney university writing a thesis on GPGPU and implementing various algorithms in CUDA.

My main goal at the moment is to write a fairly detailed paper on the architecture of a nvidia GPU and/or nvidia graphics card. However i am finding it fairly difficult to find information other than the good old grid/block division information.

Could someone refer me to a good paper or information base on GPU architecture?

Programming Manual is definitely the document you should start with.

I would recommend the following paper for a rough (but detailed) estimate. For your purpose, sticking to the initial architecture description should be enough.

http://portal.acm.org/citation.cfm?id=1331…&coll=GUIDE&dl=