I see the problem. It is not easy to hide the idea of an implementation on the one hand and to get hints for a solution on the other hand.
I am on your way that only some of the requirements will be met. That is my conclusion upto now. - Concerning the task: “Integration” will be done in the first step. In the second step it might be that any kind of virtualization will lead to testing “without integration”.
The idea behind validation of software for being independent from hardware points also to
- integration of the system
- together with using e.g. graphic card 1 and
- another system with e.g. graphic card 2 and
- after integration having a set of tools that run on both systems and tell you without full product validation “yes, the system will do its’ job” OR “no, there will be e.g. performance issues not working with this hardware”.