Beschreibung:
This talk discusses optimized collective algorithms and the benefits of leveraging independent hardware entities in a pipelined manner. The resulting approach uses overlap of computation andcommunication to reach this task. Different examples are given.