CUDA Driver API
The CUDA Driver API
is the userspace component of the NVIDIA CUDA drivers. It provides utilities
familiar to users of the C standard library: a cuMalloc function for
allocating memory on GPU devices,
for example.
The CUDA Toolkit. The CUDA Driver API (highlighted) sits between applications or other toolkit components and the GPU. Adapted from the Professional CUDA C Programming Guide.
Very few CUDA programs are written to directly use the CUDA Driver API. They instead use the CUDA Runtime API . See this section of the CUDA Driver API docs.
The CUDA Driver API is generally not linked statically. Instead, it is linked dynamically, typically under the name libcuda.so on Linux systems.
The CUDA Driver API is binary-compatible: an application compiled against old versions of the CUDA Driver API can run on systems with newer versions of the CUDA Driver API. That is, the operating system's binary loader may load a newer version of the CUDA Driver API and the program will function the same.
For details on distributing CUDA C applications, see the CUDA C/C++ Best Practices Guide from NVIDIA.
The CUDA Driver API is closed source. You can find its documentation here .