CUDA Driver API
The CUDA Driver API is a lower-level interface for device contexts, modules, memory and kernel launches. On Linux, its implementation is exposed through libcuda.so.
How it works
The interface lets an application explicitly manage loaded device code and execution contexts. Higher-level frameworks and the CUDA Runtime can use this layer underneath their own abstractions. The driver library must correspond to a supported installed driver configuration.
Work through an example
A runtime compiler can produce a device-code artifact, which the application loads through a driver module API before looking up and launching its kernel.
Where the term can mislead
A link-time stub in a toolkit is not the actual runtime driver. Accidentally loading a stub or the wrong library path can make a deployment fail before any kernel runs.
Connect the idea
Take it into practice
Continue with the existing concept, apply it to an interview problem, or follow a course lesson. Access requirements are shown before you open a page.
New to GPUs? Start with the free execution-model concept.
Sources and scope
Examples here are teaching calculations, not measured GPU performance. Consult the linked documentation for the exact device, instruction and tool version. Source review: 25 October 2026.