I only have a 1070 card in my testrig. With the opensource i can reach 275-280mhash with oc. My private kernel oc around 100mhz higher than the opensource so you get an additional 5% from the oc. 305-310mhash stable.
Receive a 5% increase and give 3.3% + 0.1 BTC - it's not fair.
If you have a gtx 970, 980 or 980ti card (compute 5.2 maxwell) the increase is 20-25% on the standard clocks.(release #3).
I need more time to increase the pascal speed. Next up is a speedup for the gtx 1060. I have a test card coming soon.
You can get 10% more if you know how to configure your rig properly. I have written many posts about this but I can write it again..
-Reduce the memclock (with nvinspector you can reduce -1000mhz, with a bios mod you can reduce to 850mhz or lower)You can use the nvidia-smi tool to get the supported clocks mem/core and use the lowest value for the memory
-Overclock the core to 2050-2100 or more. When you reduce the speed of the memory you free power that can be used to overclock. Since my kernel doesn't use memory. 0% load on the memory controller makes +100mhz stable oc on the core possible.