Hardware
How DOCA GPUNetIO Unifies GPU-Initiated Networking Across the NVIDIA Software Stack
DOCA GPUNetIO delivers a unified GPU‑directed networking (GDA‑KI) foundation that lets CUDA kernels drive Ethernet, RDMA, Verbs and DMA operations without the CPU entering the application path. It shi
DOCA GPUNetIO delivers a unified GPU‑directed networking (GDA‑KI) foundation that lets CUDA kernels drive Ethernet, RDMA, Verbs and DMA operations without the CPU entering the application path. It ships both as a full DOCA SDK and as a lightweight open‑source Verbs implementation that dynamically loads closed‑source functions; this shared backend is now used by NCCL GIN (v2.27), NVSHMEM 3.7, and NVQLink’s RoCE Transceiver, providing better scalability for small messages and achieving ~2.6 µs round‑trip latency on IGX Thor/Blackwell hardware. See the DOCA SDK GPUNetIO programming guide and the open‑source repo for examples and implementation details.
Related
- NVIDIA BlueField-4 Powers New Scale-In Network Infrastructure for Agentic AI Factories
- Build Applications on NVIDIA BlueField Faster with NVIDIA DOCA Agent Skills
- Expanding AI Storage Access with NVIDIA cuObject and the NVIDIA SCADA Server SDK
Source: NVIDIA Developer | 2026-10-06