I need the detailed data of RTX3060, especially the detailed data of SM, so that I can more reasonably divide threads when programming in CUDA
There isn’t a separate white paper for RTX3060. You should use the white paper for RTX3090, and the principal difference will be in the number of SMs. The detailed data of the SM should be the same. There are other differences such as memory interface, L2 cache,etc. but these are not per-SM data, and to a first order approximation they don’t affect the discussion of “divide threads”.