NVIDIA Working On New GLX Extension To Help PRIME GPU Offloading

NVIDIA has been working on a new GLX extension to help the PRIME GPU offloading situation where multiple GPU vendors are involved and thus different OpenGL driver implementations. In particular, the proposed GLX_EXT_server_vendor_select is designed to better fit in to address PRIME GPU offloading obstacles introduced by their work on the now common GLVND OpenGL Vendor Neutral Dispatch library.

"For GPU offloading in libglvnd, where individual clients can run with an alternate GPU and client-side vendor library, we'd need some way for that alternate vendor library to communicate with its server-side counterpart. Normally, the server's GLXVND layer would dispatch any GLX requests to whichever driver is running an X screen. This is a GLX extension that allows a client to tell the server to send GLX requests to a different driver instead," explained NVIDIA's Kyle Brenneman in the proposal sent last week.

The fundamentals are keeping a separate mapping for each client in order to keep track of the respective vendor libraries and matching the GLX requests to the respective server-side driver. The proposed extension is GLX_EXT_server_vendor_select and is outlined in the aforelinked proposal.

We'll see where this leads for improving the PRIME GPU offloading situation in the GLVND world but so far there hasn't been any other developers commenting on the proposal either via the mailing list or the new GitLab merge request.