Webinar 2020 Collaborative Groups in CUDA

Jump to navigationJump to search

Starting with CUDA version 9, collaborative groups were introduced. These permit synchronization of groups of threads within a kernel, not necessarily just within a single block of threads, but at sub-block and multi-block granularity. With this, collective operations on such groups become possible. This seminar will discuss collaborative groups and how they can be used to improve your CUDA code.