Quote from: salfter on July 15, 2017, 08:24:34 PM

Quote from: fullzero on July 11, 2017, 10:16:22 PM

saflter your newest version of switch was causing problems when run with a monitor connected (LOCAL); I would recommend relying on the:

IAmNotAJeep_and_Maxximus007_WATCHDOG

to handle miner crashes / 0 hashrates.

I spent a couple hours testing this, and it is very effective; it is worth noting that it currently only works when the mining process is launched in a screen ( I will make it work for all the clients even when run locally soon: so don't spend a lot of time upgrading rigs with this)

I've been laying low the past few days, doing stability checks. Unloading/reloading the driver while switching hasn't completely eliminated crashes, so I'm thinking that a watchdog process would work better.

My latest commit comments out the GPU-resetting code, and also logs whenever it switches coins to algo-log. I definitely need to look into the existing watchdog code, as it seems that avoiding crashes is nearly impossible. I'm not even overclocking my cards, except for one algo (daggerhashimoto) where the GPU is underclocked and the memory is overclocked. It seems that if a GPU is going to hang, it nearly always does so right after an algo switch. If it doesn't hang then, it'll most likely run indefinitely.

I have been testing different values / versions of the watchdog; and have found that the version on v0018 needs to be relaxed / slightly modified. When this is done; the watchdog and your switcher work together well.