Two issues here:
1. Qubes needs to continue using its own graphics after passing a
(secondary) graphic card to a vm. If your system has two GPUs, you
cannot pass both to a vm.
2. Graphics cards/GPUs often employ proprietary tricks and shortcuts on
the PCI bus, and IOMMU passthrough is not one of the situations the
designers intended to use those tricks... so compatibility suffers.
Nvidia is probably the last brand I would expect to behave under unusual
circumstances... I would consider it a poor choice even for regular
Linux use.
Assuming there is some chance it could work, you would want to assign
your integrated GPU as the primary graphics adapter (probably in BIOS
settings) and pass only the Nvidia GPU to the vm. You may also have to
blacklist the Nvidia GPU in Qubes dom0.
Check out the other threads such as this:
https://groups.google.com/d/msgid/qubes-users/eb5bc2c6-cb40-432c-94cd-7dcc49b83b95%40googlegroups.com
Chris