GPU development in Python 101
summary
Claims: Python can write GPU code no C++ required and run it on Colab T4s. Tools actually used: Numba kernels, cuPy, cuDF (the cuDF pandas plugin), Dask with CUDA and NVML/nsight for introspection. Concrete numbers: a Python loop takes ~11 s, a naive GPU dispatch ~0.3 s, explicit device-managed runs drop to tens of ms or even microseconds. Built examples: fizzbuzz kernel, image blur and grayscale kernels; they execute and expose memory/dispatch and interoperability via the CUDA array interface and Dask + CUDA integration.
AI-written summary · about
chapters (18)
- 0:00Welcome to GPU Python
- 1:09Setting up environment
- 4:02GPU versus CPU computing
- 8:32Numba for GPU programming
- 13:00First Numba lab
- 21:47Numba CUDA kernels
- 33:20Optimizing GPU memory
- 44:20Break and lab
- 50:48Multi-dimensional indexing
- 1:01:02Image processing lab
- 1:02:25Advanced topics overview
- 1:03:51Monitoring GPU performance
- 1:11:03GPU library interoperability
- 1:24:50CuPy acceleration lab
- 1:35:11RAPIDS cuDF intro
- 1:50:29Q&A on RAPIDS
- 2:02:45Scaling with Dask and multi-GPU
- 2:13:51Closing remarks
related talks
02h 28mGPU Accelerated Python
48mGPU Programming in Pure Python
34mAccelerating GPU Algorithms in pure Python
29mA Beginners Guide to GPUs for Pythonistas
33mWhen GPUs Make Python Slower: Patterns and Pitfalls
35mCUDA in Python- A New Era for GPU Acceleration
35mPython at the Speed of Light: A Complete Tour of the CUDA Ecosystem
30mCUDA in Your Python: Effective Parallel Programming on the GPU