We are doing simulation (and collision response) on GPU using compute shaders. The GPU simulation works fine in multiplayer… but only on listen servers.
Dedicated server seems to be compiled without DirectX and without the render thread (all ENQUEUE_UNIQUE_RENDER_COMMAND macros are empty). Unfortunately this also means that no shader programs are loaded and no compute shaders are executed.
Is it possible to configure the server build in a way that:
- The server initializes DirectX (for DirectCompute)
- The server loads shaders
- The server runs the render thread (making ENQUEUE_UNIQUE_RENDER_COMMAND macros work)
I don’t need to run any viewport rendering code (no visualization). I don’t need to load textures, meshes, etc on the dedicated server.
If this is not possible out of the box, what would be the best way (with realistic work efforts) to create a custom hybrid configuration between the dedicated and listen server that meets our requirements?