There is a difference between performance mode and performance levels. Performance mode  0 is "prefere maximum performance" mode 1 is "adaptive" and 2 is "auto". Performance level is the thing with the clock speeds(adaptive clocking)1 to 4, witch you can see in the gui. It looks like a by design "feature" from the nvidia driver. Everytime you start the miner software the performance level goes to 3 and the card clock will go down. Overclocking settings will also effect the clock speed on a lower perfomance level.
Example : level 1   200mhz  and 400mhz
                  level 2   400mhz and 100mhz
                  level 3   1200mhz and 3000mhz
                  level 4    1700mhz and 4000mhz
After overclocking gpu +100 memory +1000 it will look like this:
                  level 1   300mhz  and 1400mhz
                  level 2   500mhz and 1100mhz
                  level 3   1300mhz and 4000mhz
                  level 4   1800mhz and 5000mhz
The only thing you do with overclocking, is making your card clocks speed as fast as it should be.
It would be super if someone has a solution for that.
Btw nvidia-smi -ac 5000,1800 won’t work in my case.(application clocking)

There is also a post at the nvidia devforum with the same problems.

Hi guys,

While I use v18, I get this error.



The vga params are the same as in v17 (7x1070, power limit 100, proc: -200, mem: +1100). In v17 it was stable as rock for weeks.
The problems:
- I use watchdog param but it is not restarting
- once restart after 8 hours
- ctrl+c does not work
- in the first 10 minutes I can reboot from console
- I can not use teamviewer

Please help me to find a solution.

Thank you ind advance!

could I also get this answer? just having the same stupid issues with any pools that require workers. if you prefilled the bash with your own working worker info and stuff it might help idiots like me get started. I did notice that I got an authentication stratum error when just trying to run with your prefilled LBC information