I see around the web in CUDA developer communities, when it comes to "texture cache", people are talking about "block linear". Can anyone please clarify and explain this term?
Block linear is a (not so often used) term referring to the opaque internal memory layout used in CUDA arrays. This is ordered ("block linear") to perform better in the texture cache, where spatial data locality is typically better suited to the applications which the texture and surface hardware is designed for (i.e. rendering).