They've been massively rewriting chunks of code for the driver meaning a lot of my work hasn't yet hit the firmware. I've just finished refactoring the first solid working version of the autotune for these which is working pretty well now so hopefully you'll get to use it in a new firmware soon. It takes about half hour to tune and will be running by default in future versions; it tunes every startup. If you're looking for more efficiency, this should help provide it 
I haven't realised any dramatic hashrate improvements yet though.
-ck, I have to ask you: why don't you do the autotune correctly? I haven't realised any dramatic hashrate improvements yet though.It is well known that autotuning at reboot is patently wrong: the device is probably cold and seriously out of the thermal equilibrium. It is well known that that type of tunning needs to be done repeatedly as the system reaches equilibrium. I don't have practical numbers for mining farms, but I presume most of them aren't working exactly in temperature-stabilized conditions, so thermal equilibrium will probably change at least twice per day. I doubt that 30 minutes is enough to reach the normal operating temperature from the cold start.
Is this a problem with storing the calibration coefficients in the flash memory that will wear out? Then why not store them in a circular log file with fixed-size records that are as long as the write-blocks of the underlying storage medium? Those are fairly big 64kB or 256kB or some such, there's plenty of room to write some calibration results. And if writing at every calibration is too often then write the summary (average, minimum, maximum) once per day.
Or was the "autotune" simply a marketing requirement or a contractual agreement? I kinda don't want to see the repeat of what had happened to Spondoolies and their ill conceived POST (Power-On Self Test) and how people had to resort to preheating their miners with hair dryers.
