NVIDIA announced DOCA GPUNetIO, a GPU-centric networking SDK layer that enables CUDA kernels to directly drive Ethernet, RDMA, Verbs, and DMA operations while keeping the CPU out of the application critical path. This framework is designed to unify GPU-initiated networking across the NVIDIA software stack, benefiting multiple communication libraries.
Key facts
| Fact | Detail | The source says |
|---|---|---|
| Purpose | Enables CUDA kernels to directly drive Ethernet, RDMA, Verbs, and DMA operations | “DOCA GPUNetIO provides a unified GDA-KI foundation that enables CUDA kernels to directly drive Ethernet, RDMA, Verbs, and DMA operations…” |
| SDK Version | Full DOCA SDK superset | “The framework ships as both a full DOCA SDK superset and a lighter-weight open-source Verbs-focused implementation that can detect the SDK…” |
| Open Source Version | Lighter-weight, RDMA-Verbs-focused implementation | “NVIDIA also publishes an open-source GPUNetIO project as a lighter-weight, RDMA-Verbs-focused implementation designed for frameworks that…” |
What happened
NVIDIA introduced DOCA GPUNetIO, a GPU-centric networking SDK layer designed to enable CUDA kernels to directly drive Ethernet, RDMA, Verbs, and DMA operations while keeping the CPU out of the application critical path. This framework addresses the need for networking and data movement to behave like first-class GPU-controlled operations rather than host-driven services. DOCA GPUNetIO provides both CPU functions on the control path and GPU CUDA functions on the data path, allowing the creation of CUDA kernels that can manipulate transport objects exported with DOCA Ethernet, DOCA Verbs, DOCA DMA, and DOCA CommChannel. NVIDIA ships GPUNetIO in two forms: a full DOCA SDK version and an open-source, RDMA-Verbs-focused implementation.
What to weigh
- The framework is available in two versions: full SDK and open-source.
- The open-source version is a subset of the full SDK version.
FAQ
What does DOCA GPUNetIO enable?
DOCA GPUNetIO enables CUDA kernels to directly drive Ethernet, RDMA, Verbs, and DMA operations while keeping the CPU out of the application critical path.
What are the two versions of DOCA GPUNetIO?
The framework ships as both a full DOCA SDK superset and a lighter-weight open-source Verbs-focused implementation.
Which communication libraries benefit from DOCA GPUNetIO?
Multiple communication libraries including NCCL, NVSHMEM, UCX/NIXL, and Holoscan Sensor Bridge now build on this shared GPUNetIO foundation.
Source: NVIDIA
