Thanks for attempting that experiment. This does seem to support the idea of a regression in the core DisplayPort code between 460 and 470 on this particular system, with the symptom being rather more dramatic on Windows.