It sounds like you need to be a little more engineering oriented about this.
Capture all the system specifications – OS version, GPU kind, OS RAM, GPU RAM, Driver date, display size – into a spreadsheet.
Sort through that spreadsheet, looking for commonalities.
Also: When they get the error, what is the actual stack trace?
If you go through the code, looking at that call chain, can you tell what’s going on and what the code is trying to do?
Given that knowledge, can you come up with some hypotheses for what could be going wrong, and add logging to prove/disprove each hypothesis, and then try again?