Now I have one mining and the other mines for a few minutes then switches to a gazillion gigahash and 100% HW errors
I had one like this. Well, the high GH and 100% errors part. But it would usually go for days between random occurrences. Cycling power always got it working again, usually for another few days. I sold it to plasticuser recently, so keep an eye out for any solutions he finds.
I reflashed it to 14e. It does the start-up LEDs, BFGminer recognizes it, it then reports it's doing TH/s, 100% error rate, and stops. In a windows VM, the chiliflash app reports it has four working chips: two with 11 engines and two with one engine. The total theoretical speed varies from 2 - 4GH. It's something hardware related, I believe.
Shame I only got about 18 hours of hashing with it before it fell over, but those are the breaks. I wish I could tell which ASICs were good.
Now BFL has sharply reduced prices on their 65nm line, I'm open to buying another chili or two if the price is reasonable. I'm not put off by one fussy board.
I've never seen anything like that. I had one that came in that had one ASIC that didn't work; it had a crack along one side of an ASIC and another ASIC had a crushed corner. That was pretty obvious physical damage though.
It's too bad that the ultrasonic bath didn't help any further. Do you know if Keefe used something conductive/capacitive like Arctic Silver?