This 42-minute conference talk from NVIDIA GTC 2025 explores scaling applications beyond single GPU capabilities using various multi-GPU communication libraries. Learn how to compute larger workloads or achieve faster performance when dedicated multi-GPU libraries aren't available for your specific needs. Discover the practical applications of CUDA-aware MPI, NVSHMEM, and NCCL through real-world examples presented by Jiri Kraus, Principal Developer Technology at NVIDIA. The session (ID: S72578) focuses on accelerated computing libraries within the Models/Libraries/Frameworks topic area, covering NVIDIA technologies including CUDA, CUDA-X, NCCL, and NVLink/NVSwitch. Perfect for developers looking to expand their applications across multiple GPUs and nodes for enhanced performance and scalability.