GPU Compute Support in Reducers/Procedures

[themohitkhare]

I’d like to propose adding support for GPU compute access within SpacetimeDB. Currently, application logic (Reducers) runs primarily on the CPU. While this is incredibly fast for standard relational operations, certain real-time tasks would benefit significantly from hardware acceleration.

As SpacetimeDB moves toward becoming the "one-stop shop" for backend logic, developers are hitting bottlenecks in areas that traditionally live on the GPU:

AI & LLMs: Running local inference for NPCs directly within the database logic without having to hop to an external service.

Complex Physics: Offloading large-scale spatial calculations or fluid simulations for multiplayer worlds.

Procedural Generation: Generating high-fidelity terrain or assets on-the-fly where parallelization is key.

Impact:
This would solidify SpacetimeDB as the ultimate engine for "Simulation-Heavy" games and AI-driven real-time apps, removing the last reason to maintain a separate "compute" server.