The attached example shows the same pixmap twice, once with the default-deduced DPR value (i.e. the DPR value set on the pixmap as displayed on the left side of the screenshots below) and once with a DPR value taken directly from the window (cf. right hand side, where we overwrite the DPR value of the pixmap).
When using anything older than 5.14, or even using 5.14 and then setting any of the environment variables that trigger DPR scaling, the result will look like in the attached images. Note how the pixmap on the left is always unsharp.
What's really odd is that when I run the example against 5.14 rc 1 on windows with a hidpi screen, then the images look sharp but will use the smaller pixmap with dpr 1. This is unexpected, it should use the @2x pixmap with DPR 2, since that should fit into the QSize(250, 250) target area.