NVIDIA GPU Drivers
The software that mediates all interaction between host programs or the host operating system and an NVIDIA GPU.
The NVIDIA GPU drivers sit between the GPU and everything on the host, both application programs and the operating system. Applications reach the drivers mainly through two interfaces, in order of how commonly they're used: the CUDA Runtime API and the CUDA Driver API.
Key idea The driver is the only component that talks to the GPU directly; every other piece of the CUDA stack goes through it.
NVIDIA has published the source for its Linux Open GPU Kernel Module.
Related#
- nvidia.ko — the Linux kernel module at the driver's core
- CUDA Driver API — the userspace part of the drivers
- CUDA Runtime API — the higher-level interface most programs use