CUDA Runtime API
The CUDA Runtime API wraps the CUDA Driver API and provides a higher-level API for the same functions.
The CUDA Toolkit. The CUDA Runtime API (highlighted) wraps the CUDA Driver API to make it more amenable to application programming. Adapted from the Professional CUDA C Programming Guide.
It is generally preferred over the Driver API for better ergonomics, but there are some small caveats around control of kernel launches and context management. See this section of the CUDA Runtime API docs for more.
While the Runtime API may be statically linked, per Attachment A of the NVIDIA CUDA Toolkit EULA , it does not have to be. The shared object file for dynamic linking is usually named libcudart.so on Linux systems.
The CUDA Runtime API is closed source. You can find its documentation here .