Just to follow up on the previously stated cold reset problem, I updated to the latest libs and still experience it (I tried with IDE 1.6.9 and latest 1.8.3). If I power off then on the arduino, it won't display anything on my matrix which is an FC-16 type (maybe occasionally flash some blocks at PON but that's all). I have to reset it and then it will work again. Sometimes when I modify something it won't even work at all.
Additionally, (may or may not be related) I added a DCF77 lib to sync with the Frankfurt DCF time signal and when it syncs, the display freezes (stops scrolling) and resumes after 1 second but it fails to update. I use this library on 2 other OLED clocks I made and it works perfectly so I am sure this library is ok. Apparently when the coherent timesync is detected, it fails to enter the routine and just skips it (I set a flag when the sync is received and update the nvram in my RTC which is never updated). The lib I am using is by Thijs Elenbaas (http://thijs.elenbaas.net/2012/04/arduino-dcf77-radio-clock-receiver-library/) FYI. I will try using the SetSyncProvider to see if something changes.
After I updated to 1.8.3 I can no longer get the display to work unless I remove the DFC lib from the code. Additionally if I reduce the blocks to 4 (I am using 5 at the moment) it works, but only at reset not at cold boot. It seems memory related (am using 27284bytes with DFC lib included and 624bytes of RAM so it shouldn't have memory issues) but I can't put my finger on what is going on. The code is based on this which I updated to work with recent IDEs and added the relevant DCF support:
http://www.instructables.com/id/Arduino-Parola-Zone-Time-Msg-Display/
I can paste the whole code here if someone wants to try it - to reproduce the erratic behaviour you need at least the DCF and RTC libs to be included else it works fine apparently.
** Update: I just ran the Parola_Zone_Display demo and that also fails to work when I cold boot the arduino (sometimes it doesn't even run on warm boot as with the above clock code). I have tried now on a brand new NANO and the same happens (tried on an UNO and 3 NANOs). However the Parola_Display example works fine (I have 5 modules) so the library is set correctly. Apparently things start acting up when using ZONES and having LESS than 6 modules. The MD_MAX_72xx test works also fine.
*** Update2: I managed to isolate the problem to minimum modules in a zone and number of modules: it seems that with less than 6 modules the code would randomly work. So I increased the number of modules to 6 from 5 and shifted the zones 1 module so that it would display correctly still on 5. Apparently now it will cold boot correctly at PON. I did the same with the Parola_Zone_Demo reducing the number of zones to 2 editing the code reducing the array to 2 blocks instead of 3 and increasing the modules to 6 and apparently this makes it work (even if the letters in one zone are shifted out due to a missing module).