Skip to main content
NETWORKING · EDGE AI

What is RoCEv2?

RDMA over Converged Ethernet, version 2 — the standard that brings low-latency, low-CPU-overhead data transfer to the routed Ethernet networks most industrial and edge AI deployments already use.

01

What RoCEv2 is

RoCEv2 (RDMA over Converged Ethernet, version 2) is a networking standard that brings RDMA capability to routed Ethernet networks, enabling low-latency, low-CPU-overhead data transfer on the standard infrastructure most industrial and edge AI deployments already use.

RDMA originally required specialist fabrics such as InfiniBand. RoCE made it possible to run RDMA over Ethernet, and RoCEv2 extended that capability to routed IP networks, which is what most industrial and OT environments actually are.

02

What changed from RoCEv1

Compared to RoCEv1, RoCEv2 made three changes that matter in practice for industrial deployment:

  • //
    Runs over UDP/IP Making it routable across standard Layer 3 networks, where RoCEv1 was confined to a single broadcast domain.
  • //
    Works with standard Ethernet switches Provided they support the necessary quality-of-service features (PFC and ECN). No specialist fabric required.
  • //
    Scales into multi-subnet deployments Suitable for larger edge AI architectures than RoCEv1 could support.
03

What to check before specifying

For edge AI, RoCEv2 is the practical way to deliver the CPU-efficiency and latency benefits of RDMA without redesigning the physical network. The main prerequisite is that the network interface cards, switches, and operating system all support RoCEv2 cleanly — a specification point worth checking during platform selection rather than discovering after procurement.

For a fuller view, see our pillar article: Why edge AI is outgrowing 2.5GbE.