Can you confirm that you are seeing v3 kernels loading in the log file? (it should be in the first few pages of the log file, immediately before the card starts to actually mine). We have tested on GTX1080, which should be the same as P104 but perhaps the better memory timings in P104 cause the difference in behavior.
yes, there was v3 kernel in log file after creating DAG. P104 is more likely 1070 with fast memory.
I found that decreasing power limit reduces hasrate on
0,038MH/1W. but I do not tested yet if it works only with v3 kernels.