Quote from: soy on September 21, 2014, 05:57:28 PM

Quote from: soy on September 21, 2014, 02:18:13 PM

Quote from: soy on September 21, 2014, 12:32:10 AM

Interesting. The problem hashing board now in my first S3 is working fine and very stable. I just found there is a difference between my first well running S3 and the S3 that the problem board came out of, and that's is a different firmware on each. I haven't found a way to determine while ssh'd in, e.g. opkg list | grep antMiner comes up a blank. But, bruno033 pointed me to the firmware page and I see there the 0721.bin has a solid green light when running. The first S3 from the mid-west has that. When I saw that the S3 from Florida had a flashing green light when hashing I though I probably configured it differently than the first. But it must mean the S3 I got from Florida has the 0711.bin firmware because the other later options are the 0811.bin and the 0826.bin, both of which will have the Advanced Frequency change option on the miner configuration page. So, the questionable board is very stable at this point in the S3 having the 0721.bin firmware. Time will tell if it remains at 441 overnight as the humidity rises, e.g. right now the rh is 30%.

soy

Disappointment. The problem board now in a good S3 this morning is down to 431.45GH/s after 22 hours. Will swap boards to their originating S3's today and upgrade beyond the OpenWrt Barrier Breaker r38031 / LuCI Trunk (svn-r9909) firmware.

Just now have both S3's back up and running with their original boards. The fast fan good board from the good S3, I should note the hashing boards had QC Passed stickers the later S3 from Florida didn't. The good board with a fast fan from the good unit which brought its fast fan rate to the questionable unit when moved yesterday, I took apart and reapplied new thermal compound. I found the temperature sensing IC to be in the center between the two rows of ASICs, an LM75A. So running lines of compound along the top ASICs and again along the bottom but not the center wold tend to prevent the heatsink from cooling the device and result in faster fan rates. The surface mount device below the fan connection and above the upper left ASIC is a 25.000 crystal. This good board had excellent compound. Only worked on one of the two as I've run out of compound. Of the board I put back in the questionable unit, I cleaned the ASIC pins but ran out of cleaning fluid so reassembled with new paste. The fan to the questionable board still remains relatively slow. I'd like to see that higher. I'll wait to see how these settle out then upgrade the firmware.

The good board from the good S3 not only had a QC sticker but also had handwriting on the right side of the board between the ASICs, an "A" and another "A" or perhaps one was an A1. The questionable board had what looked like an "S" and a lower case "o". These letters were vertical after the fashion of oriental script.

The good S3 came up to 441GH/s avg within an hour while the questionable S3 from Florida was only hitting 100GH/s. Not one with a whole lot of patience, I decided to install the 0811.bin firmware on the questionable S3. Did. Then worried a bit as it was only doing 10GH/s avg for some time. I restarted cgminer (/etc/init.d/cgminer stop, /etc/init.d/cgminer start) and there was no apparent improvement. Then I noticed there was no frequency listed on the status page so of course into the new Advanced tab on miner config, I specified the default and saved. Up 6 minutes now and in the 400's GH/s. Time will tell.

...Now over an hour and the S3 from Florida, having the 0811.bin firmware, seems to have settled out at 435.65GH/s. What would explain this and its low fan speed is if an ASIC near the LM75A has some engines turned off - perhaps it had overheated at some point and damaged one. I have noticed that the lower center ASICs under the LM75A tend to have drier thermal paste than other ASICs. Center of the hashing board. The curve of the inside heatsink fins and the direction of spin on the fans seems the reason one board, the top chain, gets hotter and has the faster fan speed.

If an ASIC near the LM75A is damaged and turns off engines, it gives the system an inaccurate assessment of the board's temperature. ASICs toward either end working properly will be running hot.

...their logic that the lower hashing board should be on the cooler side made sense so it's been torn down again, end plates reversed, fans reversed, controller board moved to the other end, and reassembled. Now coming up.

Florida, land of the floating chads.