Over at the Nvidia Developer Zone, Mark Harris looks at how to efficiently access device memory, in particular global memory, from within kernels. Global memory access on the device shares performance ...