CUDA

What CUDA is

CUDA provides the programming model and libraries that let frameworks use GPU parallelism efficiently, and its maturity is a major reason NVIDIA hardware dominates AI.

How it works

Frameworks call optimised CUDA libraries for matrix operations and attention kernels, with custom kernels written where standard ones are too slow.

Why it matters

The software ecosystem, not just the silicon, is what makes switching hardware vendors difficult.

Common uses

• →Training and inference acceleration
• →Custom kernel development