CUDA Peer Access allows one GPU to directly read from or write to another GPU’s device memory without routing data through the CPU or host memory. It is enabled explicitly with cudaDeviceEnablePeerAccess() and requires compatible hardware (devices on the same PCIe root complex or connected via NVLink).
Direct GPU-to-GPU transfers reduce latency and free the CPU from mediating inter-GPU communication.