Odd 1070 overclocking behaviour

[LumpyVR]

Ok a question for all you overclocking guru's out there

I was having a little play around overclocking my MSI GTX 1070 Gaming X and having some odd results when using the voltage/frequency curve in MSI afterburner.

For example, if I do a straight conservative overclock of +100 on the core and memory I get a nice little boost in Valley benchmark, going from a score of about 3515 at stock (sits around 1923Mhz at 59degrees) to 3630 (sitting around 2012Mhz at 61-62 degrees). This is using a custom fan curve and stock volts.

So far so good!

If I switch the trying exactly the same using the voltage/frequency curve in afterburner, I can get it to sit at exactly the same 2012 Mhz, on exactly the same voltage (around 1050mV) but the valley score drops to pretty much the same as it was stock.

I played around upping the core voltage a little but I don't think that's it as it power shows it never getting over about 70% so its weirdly annoying!

Can anyone give me an explanation as to why I don't get the same results using the two methods? Obviously I'm assuming afterburner is doing other things with voltage and whatnot in the background but it would be nice to know the differences, and what can be done in the voltage/frequency curve to get the same performance which would then allow for maybe even undervolting a little to help temps.

Cheers!

 

[LumpyVR]

Ok a little update after playing some more. The difference is a lot less than my first results, not sure if something else was going on with my PC at the time but another few runs to compare and it came out about 30 pts in Valley, still lower but not the 100 or so pts I was seeing. Still a clear difference between the two, would be nice to know why.

EDIT - Tried the same frequency at lower mV and the score goes down consistently, so even though it reports to all monitoring tools that's its running at the same 2012 MHz, clearly its not performing like it. Is there something I can tweak to improve the results at lower mV? Otherwise the standard method of adding MHz to the core clock appears to get the best results, which I was under the impression wasn't supposed to be the case anymore with the 10 series and GPU Boost 3 etc...