My card's power limit is raised a lot with liquid metal. (I really don't recommend doing that though unless you use something to ensure containment of the LM. Soldering different value shunt resistors is a safer idea as the liquid metal can "absorb" into the solder and then you have to replace the pads on the board) That doesn't mean it will be stable though, lower voltage points on the curve may not be stable at the increased offset even if a higher point is - still that should get you reasonably close. I mean technically if you wanted to go full overkill you'd want a 1+ hour run that manually tested more/all points. Imagine you could have rapidly varying loads from things like OCCT (with error checking), mining algorithms, BOINC, and the OC scanner test. PLUS quickly switching between voltage/frequency points to expose instability from the VRM not being able to supply voltage fast enough to accommodate the sudden spike in load at that frequency. You could use OCCT intermittently to artificially increase heat as you test a higher frequency that OCCT can't because of power throttling. Actually a version of OCCT error checking that doesn't load hard enough to power throttle cards would be better. You could manually lock the steps to check them yourself... man that would be tedious.