This has been kicking my ass lately as I am trying to build a generic GL abstraction over the past year or so. OpenGL is in such an odd state that the actual graphics hardware hasn’t changed significantly in a decade or more, but the API certainly has grown to be more elegant. OpenGL 2-3 had a nice way of doing things, 3.3-4.1 is a weird and inconsistent API, and 4.3+ is nice again. But good luck finding support for 4.1+ anywhere. Not on macs, not on old cards, not on web, not on mobile.

So in 3.3-4.1 land, you get to do the exact same things you’d do in 4.3 land, but in more tedious ways programmatically, because the drivers and implementations decide how elegant you are able to make your code and abstractions. The commands that end up on the GPU are the exact same regardless but the developer experience is highly variable.

My point is GL versions do not delineate features nor performance enhancements past a certain point, they delineate periods of programmer happiness. It is solely up to the drivers, not the hardware, not the actual machinery, to determine this metric. GL, being a standard rather than an implementation, suffers for this. I happily await webgpu finalization.