I have a 2017 Dell XPS 9560 with a discrete nVidia graphics card in addition to integrated Intel graphics. The nVidia card is usually powered down by way of an ACPI command. I do not have, or want, the propietary nVidia drivers on my main system. However, I would really like to use CUDA. Can I do GPU passthrough to a KVM guest runnining the proprietary drivers?
Most resources I've found seem to assume that the host is running the proprietary drivers and Bumblebee, but I haven't been able to assertain whether this is necessary or not.