Quote from: beavertank on November 30, 2013, 11:43:01 PM

Quote from: console_cowboy on November 30, 2013, 10:56:31 PM

We need a comprehensive report of the states in which people's boards are going bad. There are a ton of reports of bad boards, but little useable information.

I think that it would be helpful to Barntech's team if we could list how boards were hooked up, i.e. miniplane or not, pci-e power or not, etc. I would also think that powersupply type, with as many details as possible would be helpful. Reports of failed boards don't tell anyone much, other than failure rates.

All four boards were plugged into a miniplane and powered by PCI 6-pin connectors from a Corsair RM550 power supply (mfgr. data sheet with specs - http://www.corsair.com/en/media/cms/manual/corsair-psu-spec-table-091813.pdf). No other devices were connected to the PSU. The miniplane was firmly connected to a Spotswood open style case and a 120mm fan was blowing directly across the heatsinks, the fan was powered by the on-board connector. Each board showed green when initially powered on, all hashed without immediate error when cgminer_drillbit.exe was run without any --drillbit-options.

Each board continued to respond, and hashed without error for approximately five to ten minutes each before being stopped, when individually started with the --usb command option to determine which USB device ID was an Eight and which was a Thumb. After all USB device IDs were tried, an instance of cgminer_drillbit was started with the --usb option so only the Eight were hashing. After approximately ten minutes, the hash rate was lower and more variable than I had hoped for, so I was going to start testing each board individually with a variety of --drillbit-options to find the optimal settings for each board.

At this point two of the four boards failed to respond when tried with or without --drillbit-options and "cgminer_drillbit.exe -n" failed to list two of the boards. On investigating physically, two of the four (the nonresponsive two) were found to have no green LED on.

I then shut everything down, removed the boards from the miniplane, plugged them into 4-pin molex connectors, and powered everything on. The two LED-less boards remained green ghosts. I then returned everything to the miniplane, powered it on, and have been consistently and stably hashing on the two responsive boards (at ~19gh/s per board with default settings) for the past 16 or so hours.

I worry the miniplane may have been to blame in some way. I saw very inconsistent performance and errors with it. I also know that molex power can backfeed through the miniplane. This is somewhat troubling to me, as I have seen this cause issue in other products. I would highly recommend against using the miniplanes at this time, from personal experience. I await the developer's analysis of the situation but, I would not use the miniplane and molex pins at this time. You can make a stack using only three of the holes on the boards and the included nylon risers, as I have done, to temporarily run without the miniplane and have them all stacked neatly.

Do you happen to have a 'history' output from the different --drillbit-options you attempted? That could be very useful, particularly the last working settings.