So the problem S3 was down around 430GH/s this morning and I decided to go over its probably reworked board, the board that had grossly over-tightened screws, but would try frequency change first.
It had been down to 430 so I tried changing the frequency to check out its effect on hashrate. In /etc/config/asic-freq I changed to f=225, rebooted and restarted cgminer to get a fresh time duration. I gave it some time but it came up very slowly only to 420 or so. And it was a cool morning! All the other machines were hashing perfectly. So, I tried again, this time with f=212.5 but again it came up slow and only to 417GH/s. So, I ripped it apart again.
I was going to concentrate on its ASICs at the hot end of the reworked board, but ended up getting distracted. What distracted me was that the board had marking in black ink at the far right under the top ASIC (on this board that would be near the intake fan). Might have been Chinese or a code of some kind. So, I cleaned that ACIC and the one under it with a spray of automobile brake cleaner such that all thermal paste was removed. Must be careful with the stuff as it will mar many plastics. I sprayed those ASICs cleaning around the pins, reapplied thermal compound to both sides of the blade and reassembled. The brake cleaner removed the notation marks on that blade. When I reassembled, for a change, I put the bad board on the right side facing the RJ45 connector.
I started it up then ran to the store and back. Right now the 'difficult' S3, at a frequency of 218.75, after 1hr 20min, has an average hashrate of 443.74GH/s. The temps are both 40° with fan speeds of 2040/1800. Time will tell.
Bummer. Up now over 6hrs and the hashrate has started to drop off. Could be to an increase in humidity and that the first S3 hasn't seen a similar tailing off of hashrate because its average is over a 12hr period but probably not as it's at 442.24GH/s utility 17 while the problem S3 has tailed off to 437.63, utility 15. Experience tells me it will continue to fall and tomorrow morning may be down around 420GH/s. Temps are 40/39 and fans 1920/1620. The good S3 has 40/38 3180/1920 and the good and questionable S3's are side by side. What jumps out are the higher fan speeds of the good S3.
Too bad. 1:30AM, checking the problem S3 and its hashrate has dropped to 429.01GH/s after 9hr 55min. Since I have two S3's, I could take what I think is the bad board and swap it into the other S3 and see if the good S3 then has a falling hashrate - just to make sure which board is bad. What's wrong with that is that the other S3, purchased from a reseller in the mid-west, runs fine (presently 441.67GH/s) while the problem S3 is from Florida (and that guy wants me to remove the 1 star Amazon rating, screw him, opened the bottom of a box from a good one and put the probably returned problem in and sent it out as new). By morning the problem S3 will likely be down to 420GH/s. Wonder if it's accumulating errors in an ASIC that then shuts down engines over time. Wonder if China sends more problem S3's to Florida than elsewhere given that Florida is a problem state like Shanghai crime before 1949. I'm not a happy camper.