Thanks Form, u're you superb ! After adjustment I'm now getting 11 - 14 G ~ with almost no error , < 1%.. 
I'm getting alot of " went idle before work was sent"
I'm getting alot of " went idle before work was sent"
Yeah, i get these messages here, too - since months
Actually, changing TICK_TOTAL adjusts the timeout for one job. When the value is too big/small the result is many dupe/idle messages.
I think 20800 is near the "golden mean". It doesnt affect the hashrate as much as one would think.
I have updated the code (and the .hex) in the github-repo now, too.
Can you tell something about your frequency and voltage settings? Are you cooling the chips from topside, too?