I have built a small HDA to utilize the GPU OpenGL Vertex Cache. I use it for fast realtime SOP viewport previews. It is much faster then the Cache SOP, but it uses Alembic as a disk cache, so it is usually a dead end. Any tips how to utilize GPU Vertex Cache without .abc are highly appreciated! An example of speed comparison, the scene is attached here:
Cache SOP ~ 25 fps
OpenGL Vertex Cache ~ 45 fps
OpenGL Vertex Cache, point cloud ~90 fps
I have also prepared simple UI to render remotely on my HQueue farm.