Some corrections on how voltage on Nvidia cards actually work.
1. Voltage slider in afterburner/px1 merely permits additional voltage steps on driver v/f curve, it does not increase the actual supplied voltage per curve step by 100mv etc.
2. Driver reported voltage via GPU-Z or HWInfo is entirely different from actual voltage being set by voltage controller. Afterburner and PX1 poll the actual VC via i2c so that voltage is closer to accurate. Driver voltage is the voltage the Nvida driver requests based on power draw and driver clockspeed. The second you start touching voltages this value is completely meaningless, since i2c via classified etc overwrites this requested driver voltage.
3. LLC can massively modify the actual voltage on NVDDD at chip. You need to actually probe the rail to observe this; luckily kpe probe-it makes this trivial.
4. OCP is not going to limit you, at all, outside of extreme voltages under LN2. The amperages required to trip OCP are in the range of actual electrical shorts, AKA not even close to the 20A or so you are gonna be maxing out on <1.2V. 1.5V when the temp isn't -140 degrees after the grease unknowingly cracks on LN2 though, now that's a different story and that is why you are able to turn it off. Also if you trip OCB it will be a hard card crash, observable as a device cycle in event viewer as opposed to a driver disconnect etc from a normal instability crash. So stop turning off OCP if you aren't benching on sub-ambient on extreme volts, its doing nothing except letting your card die instantly if you ever short.
5. The dip switches are a voltage offset applied on top of driver requested voltage. So you can simply flip a dip to increase voltage higher than what v/f curve says and therefore move the frequency higher at each point without instability. While direct i2c via classified is a static target voltage, meaning the value you select what you want the end on-chip voltage to be at all times, so the voltage controller will over supply to compensate for droop based on llc, plane loss, etc.