CUDA Live: Scaling HPC with Multi-GPU Communication Libraries

Join the developers on the CUDA Communication Library team to learn more about how you can support CUDA applications on clusters of GPUs. Communication libraries like NVSHMEM enable multiple GPUs to work in parallel on large-scale AI training, simulation, or rendering tasks by exchanging data and synchronizing tasks. In this session, we will cover: - How NVSHMEM extends the OpenSHMEM APIs to support clusters of NVIDIA GPUs. - API calls to collectively launch CUDA kernels across a set of GPUs. - Live demo of NVSHMEM in action. - Live Q&A with our panel of library developers and experts.