Attaching host devices to virtual machines

By attaching various devices from compute nodes to virtual machines, you can reduce VM network latency or accelerate visualization inside a guest operating system. The following PCI devices are supported:

• Graphics cards. With GPU passthrough, you can give an entire physical GPU to a single virtual machine, while vGPU enables dividing up video RAM of a physical GPU between multiple virtual machines. You can use both capabilities, GPU passthough and vGPU, on the same node.
• Network adapters with Single Root I/O Virtualization (SR-IOV) capabilities. The SR-IOV technology enables splitting a single physical adapter (physical function) into several virtual adapters (virtual functions). Each virtual function appears as a separate PCI device that can be attached to multiple virtual machines.
• Host bus adapters. To attach HBA devices to virtual machines, use the steps described for GPU passthrough.

Limitations

• PCI device passthrough and GPU virtualization is only available on servers that support Input/Output Memory Management Unit (IOMMU). For a list of IOMMU-supporting hardware, refer to this article.
• vGPU is supported for NVIDIA GPU cards.

Procedure overview

1. Prepare compute nodes depending on a host device you are going to pass through or virtualize.
2. Reconfigure the compute cluster to enable PCI passthrough or vGPU support.
3. Create virtual machines with attached PCI or vGPU devices.