(Booth Talk from SC20)
Today, compute GPUs have become a primary enabler to accelerate a wide range of workloads. Given the rampant growth of data set sizes and computing demands, programmers have quickly reached the limits of memory and compute resources on a single or multiple GPUs managed by a single host.
In this video, Kaeli suggests some new directions for researchers to pursue to move beyond the resources of a few GPUs, while also reducing the programming burden on application developers.
