Presentation
SIGN IN TO VIEW THIS PRESENTATION Sign In
GPU Programming with Python: A Comparative Study
DescriptionPython has increasingly become a practical tool for high-performance computing (HPC), particularly for GPU acceleration. While C++ with CUDA has long been considered the standard for GPU programming, several Python frameworks now offer similar capabilities with reduced development effort. In this work, we compare GPU performance across four implementations: CuPy, Numba, PyCUDA, and CUDA C++. Benchmarking a matrix-vector multiplication on a modern GPU system, we evaluate execution time and ease of development. Our results show that Python-based approaches not only match but can surpass CUDA C++ performance, with CuPy achieving up to a fivefold speedup, highlighting the potential of Python as an accessible yet powerful option for GPU computing.
Author/Presenter
