Edit: Please let me know your VRM temps if possible... with f43, the vrms were artificially low I believe, compared to fz,fx,fy or the sapper 950 mhz bios... same load measured with hwinfo/gpu-z and kilawatt wall meter.
I agree Belkov,
Test with OCCT 4.4.0 360x360 is good or 640x480 shaders 5, dx11 error test on, fps limit 0.. the higher res freeze instead of report errors.
Try different levels of powertune and look with hwinfo or hwinfo64 what the clock speed/voltage is... example, to test the second state, I run pt 8 , because the load of occt is so high that it pulls it down to around 995 mhz (my new state 2 @1.088v- much better as the card uses it more often when pt pulls clock down past 995- saving power and heat)
For the higher mhz, you might not be able to do a real full load test because the current load goes too high for the vrm, in that case, pt 12 or so allows me to see it jump states between my max (1100) and 1000 or so.
You will see that with more load, there is much more vdroop... so with normal programs, my 1.18v, shows 1.13 or so v. With occt it goes down to 1.08v! This is one other thing that I mentioned in the past, that the real overclock is one that can run under maximal stress.
Some people are satisfied that it works in the games they play, but not many games fully stress the gpu, even when it says 99 or 100% gpu load. You can tell by looking at hwinfo current IN load, which is the 12v in load in amps. Multiply the current by 12 to get the real wattage used.
(example: 2500k @ 4.5 ghz... weak stress test can do 1.28v, but under intel burnin test, or prime95 avx, I need 1.33v!)