CUDA Techniques to Maximize Memory Bandwidth and Hide Latency

Do you want to write speed-of-light CUDA kernels? When a hand-written kernel is the right approach for the job, you want to use the best possible practices to maximize performance on the GPU. In this tutorial, we will showcase techniques to maximize memory bandwidth and hide memory latencies in our CUDA kernels, including how to efficiently use shared memory, distributed shared memory and asynchronous data copies.