OK. The logged "little buglet" version has been fine for 32 minutes - shutting it down now to try the other stuff.

I can confirm aigeezer's experience so far, and a bit more:

cgminer-4444 : 6 zombies in 1.5 hours
cgminer-allrw : 1 zombie after 9 minutes

logfiles (without --debug) here if required:
 https://dl.dropboxusercontent.com/u/44240170/logfile-4444.txt
 https://dl.dropboxusercontent.com/u/44240170/logfile-allrw.txt

And so, on to the next one...
 

Yep - There were no zombies when I had logging and --debug on with the new "bugfix" build.  May be that's the way to go.

Edit: Two more results, both no-go's:

cgminer-allwrites : 3 zombies in 38 minutes
cgminer-oh         : 1 zombie in 5 minutes

logfiles here (without --debug), if required:
 https://dl.dropboxusercontent.com/u/44240170/logfile-allwrites.txt
 https://dl.dropboxusercontent.com/u/44240170/logfile-oh.txt

Back to 3.3.1 for me then

I have a question because i dont get it to work.

I just setup some USB Erupters for mining PPC instead of PPC.

This is the command line i use right now:
cgminer -o http://stratum.d7.lt:3333 -u user -p pass

Everything runs ok but my GPU also starts to mine. What do i add to disable my GPU from mining ?

Thanks, that solved it.

Now i feel like a retard to

Thanks very much for trying all those, guys. I'll see what else I can come up with, but this will be the only way I'll be able to find it since you can reproduce it so easily.

Starting "chop" test now. The "oh" version was still running fine with log/debug on after 4.25 hours. Looks like that pattern is holding.

I love the names for the various test versions - fun to imagine what they might stand for.

Edit: just 5 minutes into the "chop" test one LED came on for a couple of seconds but it didn't go zombie. I don't think I've ever seen that before - maybe one second at the most. May not be relevant, but it was unusual.

Edit: Breakthrough? LEDs are going on all over the place but then they go out again, often after 10 seconds or more. No zombies flagged. Every one recovers so far. Run is about 14 minutes old.

Edit: AMU 5 finally went zombie about 35 minutes into the run.

Thanks a lot, it is suspicious that something's changing. Maybe we can combine some of these experiments then. More soon.

OK, I'll try cwa2 in a moment. "Chop" is still running. It only had two zombies despite all the bright lights. I could hot-swap both and it recovered. Run time about 1.5 hours.

Edit: the cwa2 test has been interesting. It ran for about 20 minutes with several heart-stopping LED incidents of 10-12 seconds each. All recovered though until one stayed on for even longer. I looked at the display expecting to see a zombie but instead saw the hashrate counter decrementing slowly. It took the poor thing about three minutes to get all the way down to zero, where it has stayed for a while. I'll see if I can hot-swap it now.

Edit: Yes, the display changed to zombie as soon as I unplugged it, the rate started to climb back up from zero and then the device got reallocated to AMU13. The test continues with no other quirks so far.