Hi Mister-e.

The PIC always starts. I know this because I have solenoids and switches and the switches always control the solenoids correctly. I think that rules out BOD or LVP issues. The 2 x 16 LCD always starts with the top row all black boxes and the lower row blank. When it starts properly, the LCD clears in about 500 mS and shortly afterwards the expected characters appear. When it does not start, the top line remains black and nothing appears on the lower line.

I have tried 100 nf, 1uf and 10 uF as the MCLR capacitor with 10k to +5. Again the PIC always starts.

The big 12 V 10 amp power supply is relatively slow to start and turning it on/off at the wall shows the LCD problem much more often than leaving the power supply on all the time and make/break the power line to the processor board. A scope shows the +5V rise time as 20-100 mS for on/off at the wall and under 1 mS for make/break to the 12 volt feed to the processor board. dV/dT is clearly part of the problem.

A web search has turned up www.densitron.com and they have a discussion on timing aspects plus 4 bit and 8 bit modes. They make much of the tight timing restrictions between the Enable, RegisterSelect, data transitions and Read/Write lines and say this is a common problem with faster processors. sadly their app note is shot with typos and they do not define the pin functions (should RS be high or low during write? - they show it as both)

I have the distinct feeling that PBP may be marginal with some brands of LCD at higher crystal speeds.

Cheers
Brian