Overview#

The omni.gpucompute.plugins extension provides an abstraction between pure compute frameworks and graphics pipeline frameworks for managing GPU device memory. It serves as a gateway that allows applications to manage GPU memory through a unified interface, abstracting over the differences between CUDA’s direct memory management and graphics APIs’ resource management systems.

The extension contains two specialized plugins that implement analogous memory management interfaces for different compute frameworks. The CUDA plugin leverages NVIDIA’s CUDA runtime for direct GPU memory management on CUDA-capable hardware, while the D3D/Vulkan plugin provides memory management through graphics APIs. This dual implementation allows applications to seamlessly switch between compute backends based on available hardware and driver support, ensuring maximum compatibility across different systems.