With memory under-clocked I get 281MHs, even with only 1 card. At mem -500 Nvidia auto adjust my core clock to ~2090mhz which is pretty high. There is no way I could get 300+?

Is cuda 8 bad? I noticed the screenshot above the guy is using cuda 7.5.

Or maybe Nvidia 367.27 drivers are slower? I use them because overclocking in linux didn't work with some recent ones I tried.