damn, that one board is hardware error hell at that frequency. with that many hw, pool side rates are going to be affected harshly. Id sell it too 😊. id offer $2300 shipped with the psu if you want, cause id be sending that board back, or looking for a replacement board.
You are "wrong" 1227 HW errors in almost 2 days is nothing. Look at the almost 0% under DiffA#. I could overclock it more and still be in an acceptable range, or put it at stock frequency and see no errors, or flash blissz firmware and tune per board for more overclock. There is nothing to be repaired. But you know all this already, nice try.
Edit: I'm honest, not stupid.
First off, I'm not calling you either dishonest or stupid, and neither would be my intentions. I'm being honest as well. My personal opinion, if it were mine, would be to contact Bitmain and see about a replacement board. They may or may not do something about it. The hard part would be if they wanted the whole unit shipped back ( I don't think I would do that as it is a lot of lost mining time).
So I'm running mine at 400 mhz with chip temps sitting at 48-51 C and pcb temps at 42C (running in southern California), and in a 48 hour period I average about 520 HW total for all 4 of my boards combined on lower diff settings (only 16.4k on the coin I'm currently mining). For some reason I do think running on higher diff settings I see a reduction in HW (and I have no clue why). At 400 Mhz I don't really see any difference in HW errors compare to 384 Mhz
What has me concerned about your 1 board is the severe imbalance of HW errors on it compared to your other 3 boards. They should all be fairly similar to a degree. Even I have 1 board on my L3+ that produces more HW errors than the other 3 boards (they all do), but it is only 2-3x as much as my lowest HW board, not 10-20x as much as the middle HW boards, or 1200x as much as your lowest board. That tells me there is something funky with the chips on that 1 board compared to the other 3. I would recommend you knock it back down to stock frequency, run it for a day or so and watch that board. My guess is that it will still produce abnormally high HW errors compared to your other three boards. You could also try running it in a cooler area and see if reducing the chip temps down normalizes the HW errors on that board.
Also, the DiffA# has nothing to do with your HW errors and pool side hash rates. DiffA# is just the share difficulty of the last Accepted share submitted to the network (block found) if I remember correctly.