Building an ARM64 AI cluster by hand on SUSE Virtualization: PCI passthrough, NUMA, RDMA and SUSE AI Factory
Three VMs with an NVIDIA A2 and a BlueField-2 DPU passed through, pinned to the right NUMA node, provisioned by one cloud-config, joined to Rancher with Multus + Canal, and running a SUSE AI Factory blueprint. Pod-to-pod RDMA goes from 1.25 Gb/s to 24.51 Gb/s once the placement is right.