Since parallel programming is highly dependent on hardware then how can we write code for multiple machines? I.e. if coding is so highly dependent on a specific machine and we are coding on a NVIDIA GPU with specific specs in mind (I assume these are look up time, stride, number of cores etc.) then how can we translate that code to a different GPU that likely has vastly different specs?
bpr
@acortes, as we will learn in this course, there is sufficient commonality between machines that most of your work on one GPU would translate to a different GPU; however, the final push of performance improvement comes from the final specialization to the specific hardware.
Since parallel programming is highly dependent on hardware then how can we write code for multiple machines? I.e. if coding is so highly dependent on a specific machine and we are coding on a NVIDIA GPU with specific specs in mind (I assume these are look up time, stride, number of cores etc.) then how can we translate that code to a different GPU that likely has vastly different specs?
@acortes, as we will learn in this course, there is sufficient commonality between machines that most of your work on one GPU would translate to a different GPU; however, the final push of performance improvement comes from the final specialization to the specific hardware.