I just got through spending most of the day messing with the jalapeno.
I made a custom peltier assembly and a temperature controller that regulated the temperature to around 65-68 F. I didn't want to go colder due to risk of condensation.
I also modified the firmware to make the same changes as ckolivas did in his, except you can just comment out in std_defs.h...
#define __RUN_HEAVY_DIAGNOSTICS_ON_EACH_ENGINE
...and it should have the same effect. I also reduced
#define __TOTAL_DIAGNOSTICS_RUN 10
...down to just 1. After these changes, I was able to get 30 engines running from 28 previously.
I also increased the voltage from 1.016v to 1.047v by changing the resistor at R11 to 7.5kohm. This resulted in the chips automatically clocking to a higher frequency. At frequency index 7, I was previously attaining 291MHz/271MHz on the chips. After the voltage increase, it was clocking at 305MHz/291MHz.
After these changes, my hash rate went up from ~8.2 to 8.3 GH/s at frequency index 7, to ~8.85 GH/s. I was able to achieve ~9.05 GH/s with index 8 (~4.5% errors), and ~9.23 GH/s with index 9 (~7.5% errors).
Unfortunately, the peltier assembly didn't seem to have much effect on the error rate. I put the old heatsink on and it handled the heat just fine. The only thing the peltier was good for was getting a high initial frequency setting. With the peltier drawing ~90w, it wasn't really worth it to continue using.
Right now I'm running frequency index 8 @ 8.93 GH/s (~2.6% errors). It is possible for me to go higher at that index, but that was when I was using the peltier. I stuck the unit in the freezer, but I guess I didn't get lucky.
Some pictures of the peltier assembly...