The EPDC crashes on these NTX boards have always been tightly linked to race conditions and timing issues.
To make a long story short, we have way less UI latency than nickel, which will exacerbate issues like these. (Among other things, Nickel suffers through Qt's inability to fully handle Y8 as a first class pixel format, which means it runs in full color at RGB32).
(This was fairly marked on the Clara, which you mostly won't be able to crash unless you run it UR @ 8bpp, because that's when the epdc has to least amount of work to do. That's something we never did in KOReader, because UR is not the canonical portrait and we don't do HW rotations; but Plato does, and had to disable it on that device because it became a crashfest. And Plato's refresh requests are a lot more fenced than ours, which should theoretically help with this).
The Libra, first of its name, happens to be in the unfortunate situation where UR *is* its canonical portrait rotation. That should have been a good news in theory, but in practice, it probably means more issues
.
On a Libra 2, installing NickelClock (which essentially tacks on a tiny refresh to display a clock *immediately* after *most* refreshes) led to most devices not even managing to finish *booting* because of the boot anim refreshes ;o).