Graham Markall

Oct 24, 2024 | developer.nvidia.com | Michael Wang |Graham Markall

By enabling CUDA kernels to be written in Python similar to how they can be implemented within C++, Numba bridges the gap between the Python ecosystem and the performance of CUDA. However, CUDA C++ developers have access to many libraries that presently have no exposure in Python. These include the CUDA Core Compute Libraries (CCCL), cuRAND, and header-based implementations of numeric types like bfloat16, to name a few.