GPU Parallel Programming By Examples - The Bandwidth of Parallel Computing

Schedule and Place

• 14:00 - 16:00, January 29th, 2022
• Engineer Cafe - Hacker Space Fukuoka - 赤煉瓦文化館 - 福岡市中央区天神1丁目15番30号

Slides

• Link

Materials for Coding Session

• Codes: Colab Notebooks

Contents

• Brief History of GPU, Big Data and Machine Learning
• Essential Paradigm of GPU Parallel Programming
• GPU Hardware and Memory Access
• Coding Session: Bandwidth of Parallel Computing (Python)

Description

This is a tutoring plus practice event for parallel programming on GPU. It’s entry-level for programmers with Python experiences. We will work under Google Colab environment, therefore no self-owned GPU hardware is required for participation.

Understanding C/C++ is better but not required. There will be optional activities of implementing the two examples above purely in CUDA C for participants who are willing to dive deeper. A self-owned GPU is required for this practice.

A brief introduction of how the tech stack evolves into the AI era will be a warm-up from the beginning, followed by some essential concepts of GPU parallel programming that are quite different from single thread programming.

We are looking forward to your participation and let’s learn by coding!

References

• Numba is an open source JIT compiler that translates a subset of Python and NumPy code into fast machine code.

Useful Links

• CUDA C++ Programming Guide
• NVIDIA HPC SDK Documentation
• CUDA Education & Training
• Caltech Course: CS179 GPU Programming

Feedback for the Event

Thank you in advance to send us feedback! Link