Beyond CUDA: GPU Accelerated Python on Cross-Vendor Graphics Cards with Kompute and the Vulkan SDK

Deep dive into GPU processing in Python using the Kompute Python framework and the Vulkan SDK. This talk covers the motivations and trends in GPU processing, as well as the high level concepts and architecture of the Vulkan SDK, as well as how the Kompute framework extends it to provide a Python interface. The talk provides a simple example that shows how to use the Python Kompute framework, and then dives into a machine learning example building a logistic regression model from scratch.

A more in-depth version of this talk can be found in this blog post https://towardsdatascience.com/beyond-cuda-gpu-accelerated-python-for-machine-learning-in-cross-vendor-graphics-cards-made-simple-6cc828a45cc3

Kompute Repo: https://github.com/EthicalML/vulkan-kompute
Kompute Docs: https://kompute.cc/

Talk from FOSDEM 2021 Python Track by Alejandro Saucedo
Sections:
Intro: (0:00)
Why parallel processing: (0:53)
Vulkan SDK Overview & Architecture: (4:16)
Kompute Overview & Architecture: (11:00)
Simple Python Kompute Example: (14:33)
Deeper Optimizations in Vulkan & Kompute: (16:45)
Machine Learning LR Example Intuition: (18:38)
Code & Intuition for Kompute Python ML Example: (18:38)
Outro: (26:44)

Видео Beyond CUDA: GPU Accelerated Python on Cross-Vendor Graphics Cards with Kompute and the Vulkan SDK канала Alejandro Saucedo