NVIDIA Adds Native Python Support to CUDA – What It Means for Developers

If you are a Python developer who has wanted to use CUDA but were deterred by C/C++, NVIDIA’s GTC 2025 announcement brings relief: CUDA now offers native Python support.

This breakthrough means Python developers can write code, call libraries, and run models on GPUs efficiently without learning C/C++, opening accelerated computing to millions of Python engineers.

Historically, CUDA’s ecosystem centered on C, C++, and Fortran, with only third‑party wrappers like PyCUDA and Numba providing limited Python integration. The new native support marks a major shift, especially as Python has become the world’s most popular programming language.

At GTC, CUDA architect Stephen Jones declared that Python is now a “first‑class citizen” in the CUDA stack, emphasizing a design that feels natural to Python developers rather than a simple translation of C.

The Python‑enabled CUDA introduces several key components:

CUDA Core: a redesigned runtime offering a full Python programming experience.

cuPyNumeric: a GPU‑accelerated NumPy alternative that requires only a single import change.

NVMath Python library: a unified interface supporting host and device calls with automatic function fusion for performance gains.

JIT compilation that minimizes reliance on traditional compilers, improving efficiency and portability.

Comprehensive analysis tools for performance profiling and static code analysis.

Beyond these, NVIDIA unveiled a new tile‑based programming model called CuTile, which abstracts computation into tiles (small blocks) that map automatically to GPU threads, aligning with Python’s array‑centric mindset and delivering performance comparable to C++.

This evolution retains CUDA’s performance advantages while making it accessible to Python developers, effectively lowering the language barrier and expanding the CUDA ecosystem to the growing Python community.

According to The Futurum Group, there were about 4 million CUDA developers in 2023, while Python developers number in the tens of millions worldwide, especially in emerging markets like India and Brazil. Native Python support is expected to attract many of these developers to CUDA.

Future plans revealed at GTC include adding support for other languages such as Rust and Julia, further broadening CUDA’s multi‑language ecosystem and transitioning it from a specialized tool to a general‑purpose platform.

Reference: https://thenewstack.io/nvidia-finally-adds-native-python-support-to-cuda/