Hi, PhoenixMiner!
Merry Christmas and Happy new year!  

It seems that Nvidia kernel optimizations for lower power consumption no longer work for last DAG epochs so it is needed to increase core clock or PL again and again to get the same hashrate.
Please take a look if it can be done again.
btw, 440 series nvidia drivers are greatly optimized in terms of irq/queue performance, producing twice less LA on heavy gpu usage. and new cuda 10.2 also.

Next problem - when mining is resumed after reaching tstop/tstart gpu quite often get stuck after few such cycles. Can resuming be done more smooth from low intensity and step by step increasing to target?
Also intensity can be reduced in advance when getting close to tstop temp.

And what about gt auto-tuning not from 0, but from defined gt values? it will take much less time and be more accurate. so there will be no need to set exact values every time oveclocking is changed for one of gpus.
it can be a problem as auto-tuning runs after each stop/resume cycle and can take longer than reaching tstop temp again)


Thanks in advance!