Collabora's Mesa EGLDevice Work To Better Support Multiple GPUs
As covered earlier this month, Emil Velikov at Collabora has been working on EGLDevice support for Mesa. These EGL extensions originally developed by NVIDIA are being pursued by Mesa developers for better dealing with the enumeration and querying of multiple GPUs on a system.
Right now there is the DRI_PRIME environment variable to allow toggling between systems primarily with two GPUs (namely, Optimus notebooks have been the main use-case) but using EGLDevice support by the Mesa drivers the matter of GPU selection for OpenGL rendering can be made by the application/toolkit developer and for other scenarios like multi-GPU systems running without a display server.
The EGLDevice support is relevant to a wide range of multi-GPU system use-cases and Collabora is also planning to develop similar GLX extensions. The patches haven't yet been merged to Mesa for EGLDevice support, but it's in the process of happening. This isn't to be confused with any SLI/CrossFire type split-rendering but is just about allowing a standardized means for selecting a desired GPU at start time.
Those interested in an overview about this work being led by Emil, he's written a blog post outlining this current effort.
Right now there is the DRI_PRIME environment variable to allow toggling between systems primarily with two GPUs (namely, Optimus notebooks have been the main use-case) but using EGLDevice support by the Mesa drivers the matter of GPU selection for OpenGL rendering can be made by the application/toolkit developer and for other scenarios like multi-GPU systems running without a display server.
The EGLDevice support is relevant to a wide range of multi-GPU system use-cases and Collabora is also planning to develop similar GLX extensions. The patches haven't yet been merged to Mesa for EGLDevice support, but it's in the process of happening. This isn't to be confused with any SLI/CrossFire type split-rendering but is just about allowing a standardized means for selecting a desired GPU at start time.
Those interested in an overview about this work being led by Emil, he's written a blog post outlining this current effort.
2 Comments