If you read the topic of the other thread, krlnx's version is doing 3.3MHASH on the gtx 1070.
Mine is doing 3850 (+16.66%). Thats why I ask the members of the forum to test and verify my assumptions.
Ok, I made a test with my palit jetstream micron memory, 100 %tdp.
My own compile cuda 7.5 x64 3.73mhs
palgin compile cuda 9 x86 3.86mhs
So you can go and work further. You gain nothing.