You mention ACCESS_VIOLATION as the main problem. Is you project c++ based? If so are you doing nullptr checks before accessing data that may not be initialized correctly yet?
Actor initialization order is not guaranteed. Perhaps your game initializes actors in random order. Things like memory cache misses and disk speed can change the loading outcome. Are you accessing actors without checking if they are ready?
Editor and published game loading works slightly different so this can also cause crashes if not tested extensively.
If you are using blueprints only then don’t forget to use the isValid node to check if an uncertain node you are accessing is ok and ready.
Have you done any bios updates lately? Maybe you have the same problem as the op of this thread regarding XMP/DOCP in the bios settings?
Edit:
You can also try running HWMonitor in parallel with the engine and observe the cards temperatures (hotspot mostly). And see if the instability may come from the card hitting a thermal limit causing instability. If so then you could try setting the fan curve to be more aggressive so that it cools the card faster. If temps are a problem then a thermal repaste could also help the problem if it’s hardware related.
I’ve also noticed that a lot of people are getting 7900xtx crashes when discord is on. It seems to use hardware acceleration that causes problems.
Other people mention to install the display drivers in “drivers only” mode (in adv options)
There are posts where people had RAM problems that were caught with memtest. It’s grasping at straws but you could do a memory test just in case. Maybe the card is getting bad data along the way from a bad part of one of your Ram sticks.
Power fluctuations could also be a thing if it’s random. If you have access to a UPS that might clear up any power dips or spikes that could cause instabilities (sometimes even changing a wall outlet can help)