It is impressive. I remember reading something related to new abstractions in chip level. The most of the GPU drivers are actually on some specific chip and it works similarly regardless of the OS. Kernel developers actually develop against this new abtraction layer and not directly against GPU. Does someone has more information about this, or was this nonsense?
It means that the drivers likely have a giant 'firmware' which is really just a closed source driver loaded into the chip, but with a nicer api for the kernel to program against
