Interesting, Ive had many pi's over the years. Not one single issue.

I had exactly the same thing with GenTarkin software.
Better use  Original KNC software with the same volts.

I did this and it is no problem any more. When i  replace the memory card to GenTarkin software ,  the problem is back.

This is happening because those particular dies are so unstable in any given point in time that multiple attempts were taken to get them hashing again through the form of soft hard resets done by my firmware.
In short it means those dies are unstable and my firmware will no longer try to bring them back online to prevent bfgminer from constantly restarting.

The stock firmware will do the same thing repeatedly w/ no positive results - soft resets to infinity. While it doesnt cause bfgminer to restart constantly it can severely hamper mining performance.

You can attempt to lower the clocks on those dies and up the volts(which is stock already and I wouldnt recommend going any higher). Also, like AmDD stated, it may be an overheating issue of those dies and its causing them to become unstable, so reapplying therm paste may do the trick.   Theres no way to tell the actual temps of the dies on these Titans.

Also, if you want you can save the log files and email them to me via my website.
Lastly, hitting "CLEAR" will unflag those dies and next time bfgminer runs it will retest them.

No, they are selected  cubes  which has never been a problem with hashing at default settings.

Odd ... well, unless there is some unknown bug in my firmware which triggers erroneous die resets, those cubes most likely do have some sort of issue going on.  Afterall, your cubes may be one of those that after many upon many software resets the dies come back just fine. There are a small number of Titans which act like this. I guess it could be due to faulty pool work connection problems too.
My firmware uses the same cube/die problem detection code(die amp threshold), it just handles the situation way differently then stock firmware.
Anywho, log files would give an insight as to whats really going on.

While I don't any KNC gear, my experience with Raspberry Pi has been quite favorable. I can also say that my sum total experience involves ZERO use of the GPIO pins. Everything I have done has been USB based one way or another. My recollection is that the GPIO pins are designed for 3.3V and NOT 5V. Might that explain anything, or do I have my figures wrong?

I''ve decided to offer my KNC Titan for sale.

Mining is going to take a back seat while I  focus on finishing my trading engine and asset management software. PM me for details if you are looking for extra cubes or a whole system with dual EVGA 1300s. Its a good one. BTC accepted and escrow is ok.

Would be nice to sell here instead of on ebay.

Well, that sounds like it should be true, but it isn't. No one knows what is going to happen to LTC difficulty and prices in the future. Titans have been profitable from day 1, and will probably continue to be. Mining with a Titan is still probably the most cost effective way of generating BTC income via mining and the Titan lifecycle looks set to beat any other mining equipment past, present or future.

Why thank you General. You are a gentleman and a scholar.

When considering the profitability of scrypt mining you must also, of course, take into consideration all the little side opportunities presented, like XPY, GAME and others which were briefly (but handsomely).

Mining scrypt with Titans is profitable. Capital costs amortized over the lifetime of the miner are less of a factor than many assume. Operating costs (electricity) is what matters most. I'd expect that most Titan owners have recovered their initial investment by now and are earning 30% a year on their capital with potential for improvement.