Canonical Plans To Integrate NVIDIA DOCA-OFED Into The Ubuntu Archive
Canonical Plans To Integrate NVIDIA DOCA-OFED Into The Ubuntu Archive
https://www.phoronix.com/news/Ubuntu-Integrate-DOCA-OFED
Publish Date: 2026-03-16 20:23:00
Source Domain: www.phoronix.com
An interesting Linux detail from today’s NVIDIA GTC 2026 kickoff is that Canonical will be integrating NVIDIA’s DOCA-OFED software framework into the Ubuntu Linux archive for leveraging the high-speed networking stack for HPC and AI.
NVIDIA’s DOCA is the software framework around their BlueField networking platform with DPUs and SuperNICs. DOCA-OFED is a subset of the full NVIDIA DOCA package that from the Mellanox period was known as MLNX_OFED.
Similar to last year’s announcement of Canonical supporting NVIDIA’s CUDA in the Ubuntu archive for making deployments easier, today’s DOCA-OFED announcement for Ubuntu is similarly about enhancing the easy deployment experience for NVIDIA’s high-end HPC/AI wares on this Linux distribution.
The NVIDIA DOCA-OFED packages will be tested with the Ubuntu Linux kernel versions and plans for a more simplified deployment experience. Canonical’s announcement argues the status quo:
“Historically, system administrators had to manage networking drivers through external installers or complex manual builds, often leading to version conflicts or kernel mismatch issues during OS updates. The new workflow will remove common operational pain points such as kernel drift, driver incompatibility, and CI breakage following kernel or OS upgrades.”
Canonical’s press release adds:
“Canonical, the publishers of Ubuntu, will integrate and distribute the NVIDIA DOCA-OFED networking driver with Ubuntu, further accelerating the enablement of NVIDIA platforms.
NVIDIA DOCA-OFED is a widely adopted high-performance networking stack, commonly used in large-scale AI factories and HPC clusters. By exposing advanced capabilities like RDMA (remote direct memory access) and NVIDIA GPUDirect, NVIDIA DOCA-OFED provides CPU offload, lower and more predictable tail latency, and sustained throughput under load. This enables ultra-low latency and high-throughput data transfers – essential for training large language models (LLMs) and running complex…
