It's not so much the Kernel Memory Dump is the important step, the others are. You can have KMD set as the desired crash type, but so long as all of the prerequisites are not in place (WER, pagefile, etc), it won't generate.
Also, it's very important to keep in mind that if it is a hardware problem, the crash dump process may not be completing gracefully, therefore all we get is a triage dump (as Jared said, a snapshot of the thread stack, loaded drivers, etc). The dump process has to completely fully and gracefully to capture the entire kernel memory space.
With that said, if the crashes weren't happening before installing the new GPU, I would revert back to the old GPU and see if the crashes stop. If they do, then there's the problem. No use in creating headaches with dump settings if you don't have to.
