Let's give this a shot. "silicon lottery" applies just like it does with CPUs, so I'm going to be a bit conservative over my own numbers to try to ensure stability for you while still trying to keep voltages low. You'll have to test to ensure stability on your particular GPU. I find it easiest to create a scatter plot of the frequency/voltage values so you can visualize what you're adjusting.
MHz / mV
590 / 850
950 / 890
1105 / 910
1175 / 930
1240 / 960
1285 / 1010
1320 / 1055
1400 / 1160 (in case you want to OC a bit, but get the 1320 validated first)
Memory voltage = 950mV (as explained in the article I linked earlier, this will set a "lower limit" to the core voltage since the core and memory get the same voltage but the memory doesn't have 7 frequency states, it's either idle or full speed. The core voltage will go below this value when the card is idle.)
And this is for brichards & King_V. My personal RX480 was running quite hot (85C @ 2400+rpm fans) even while undervolted at 1300MHz/1020mV, so I took the cooler off (4 screws), cleaned off the factory thermal paste, and applied new (Noctua NT-H1) and immediately restested with a result of 73C @ 1100-1200rpms fans on full load.
And no, the "Warranty Void if Removed" stickers are not enforceable in most countries.
It's a bit cooler in my office now that it's fall (~22C ambient), but I've had FurMark running for the past 10 minutes and I'm at 70C with 900rpm fans at 1300MHz/1020mV.
My CPU is cooled by a 120mm AIO mounted on the rear fan exhaust in my case, so I'd actually get slightly better GPU temps if my CPU were loaded also since that would cause the AIO fans to ramp up and exhaust more GPU waste heat out of my case.