0 Replies Latest reply on Jan 14, 2009 5:11 PM by kos

    Does memory need to be read coalesced ?

    kos

      I know that sequential memory access is faster than random on any gpu, are there any patterns developed to maximize memory bandwidth ? Will it be usefull to make something like CUDA's coalescing global memory access ? Does LDS devided into a banks like nVidia's shared memory ?