It does seem random. After a reboot, the issue (only top-left quarter of screen is used for stimuli drawing and the rest is green) disappeared. I am running the test script via `octave --eval` so the environment is clean every time. `Screen('Rect', w, 1)` reported [0 0 3840 2160]. Displayed graylevel seems fine. By the two files, I meant the updated PsychImaging.m file and kPsychNeedTripleWidthWindow.m file in your master branch.
I think, at this point, we need measurements to be sure that it is actually working and how good it is. I got the hardware and some template program for doing so, but am quite busy. Should be able to measure the performance within 2-4 weeks.