I just tried the newest build, and somehow the hardware errors are around 180 instead of 180000 and speeds are as good as the old version. Whatever you did worked, thanks.
I was using the -g option with various values but I find it increases rejects and for some reason -f1 decreases the speed even though its supposed to increase it for my system.
Are the grouped rejects related to a pool specific protocol like longpoll, is that what you meant?
I have more testing and pool hopping to do later, but for now I am happy with this miner again.
I seem to have fixed the problems with nvidia hw errors, but not the extreme minority of AMD users that get it (which I still believe is a very obscure driver bug).
-g is ignored when using LP.
Grouped rejects happen on all pools no matter if they use LP or not because the client runs multiple getworks in parallel, if the pool hiccups it will hit all of them at once.