kvmqemu

An overview of enabling GPU passthrough to virtual machines for AI labs: covers hardware/BIOS prerequisites (IOMMU, VT-d/AMD-Vi), hypervisor setup (KVM/QEMU, Proxmox, ESXi), NVIDIA driver and CUDA install, device isolation with VFIO, common pitfalls (IOMMU groups, reset quirks), security and resource planning, and tuning to achieve near‑native performance for training and inference workloads.