When a customer configures the NodePool with multiple or a very big HostDevice, the devices may not work properly due to lack of a bigger MMIO space. The drivers/device may fail to initialize inside the Guest VM.

This is particularly important with devices with a considerable about of memory such as vGPUs (>24G), and/or exposing multiple of them to the same Guest.

The easiest way to present the VM with the full space supported by the CPU is to use a CPU model of "host-passthrough", see RHEL-35391.

Please consider setting the CPU model automatically on the VMs created by the NodePool based on the device characteristics (and/or count), or allow the customer to specify the CPU model to 'host-passthrough' for the NodePool VMs - the option is currently missing, in nodepool.spec.platform.kubevirt.compute only cores is exposed.

Without this customers cannot use big or multiple vGPUs per node in HCP Kubevirt.