So! All units are hashing over 440gh (Two over 490), but only when using a cronjob to restart the cgminer service every 1/2 hour - 1hr depending on the miner.
Regardless of how long I leave any of these guys up to "stabilize", they always fall to 390-398 after 48 hours.
There's only one thing. I am noticing a rather high number of duplicate shares poolside since starting this restart routine. (surprise)
Is there a cgminer flag among the bitmain options to clear the asic of all work upon a cgminer restart? I know the A1 miners have a similar command with '--hwreset'
Restarting every 30 minutes is a bad practice.
Try replacing thermal paste grease and I think the issue will be solved.
I'm pretty aware how just how shitty it is, but unfortunately after re-pasting two units I saw a decrease in running temperatures but was still experiencing a steep drop off in hash after about 25-35 minutes. One of the units will spike back up to about 470 but never enough to average out to 440 without cgminer restarts.
I wish I could attribute this to poor labor skill on my part, but I'm just about 100% certain I don't suck at apply thermal paste. =/
Your miner has some series thermal issues as per my understanding.
Restarting is still a bad practice if doing it every 30 minutes. If it was once in 5 hours or so then no problem.
You average in 30 minutes is less bearable and for taking an average need to run atleast 2 hrs due to the range of move S3 does.
I advice you to findout the culprit in your miners, corner it and fix it.
Can you post your Miner Status page 5 minutes after starting S3 and 5 minutes after its hash rate is lowered, thats 5 minutes after 35 minutes.
We may try to fix if you are interested.