Details are here:
https://github.com/VirtualGL/virtualgl/issues/10#issuecomment-680390532
Conceptually, this feature brings back functionality that we used to
have on SPARC machines "back in the day" (via a proprietary Sun API
called "GLP")-- the ability to directly access a GPU without going
through an X server. nVidia introduced that same ability in early 2016,
but due to the lack of double-buffered Pbuffers in the EGL API, it took
several years, research grants from multiple organizations, and several
false starts to figure out how to fully emulate GLX using EGL. (The
final solution involves a complex abstraction layer that uses RBOs and
OpenGL context sharing to emulate multi-buffered Pbuffers.)
Please test with your 3D application of choice.
DRC