PGAP 2021 Day 1

Short course on conservation laws on GPUs day 1

Content specific for 2023

• Live video lectures from 2023
• Software installation

Content for 2021

• Day two
• PGAP course homepage

Keywords: CUDA, Python, OpenCL, C++, Git, logging, measuring performance, parallel reductions, gpu architecture, ...

Description: In this series of lectures, you will learn how to solve simple conservation laws that simulate things such as acoustic waves and heat dissipation. This is a practically oriented course where the aim is that you work through tutorials and exercises, and you need to do the exercises to understand the concepts being presented. By the end of this course, you should have a Git repository with source code that you have written yourself.

The course is structured into the following five topics (todays topics hilighted):

1. Introduction to GPU computing;
2. Computing PI with CUDA;
3. Best practices for scientific computing;
4. Optimizing GPU performance;
5. Physics simulation on GPUs.

Introduction day 1

Session 1 - Parallel computing intro

• Slides

In this video, we go through an introduction to parallel computing, and why we need to care about parallel architectures and parallel code.

Session 2 - GPU computing intro

• Slides

In this video, we go through a crash course in GPUs - with a short intro of how they work architecturally, as well as how we program them using Cuda and OpenCL with grids and blocks.

Session 3 - Computing Pi with CUDA

• Slides

In this video, we go through computing Pi (3.14159...) using Cuda. The example is to show how an algorithm gives an initial speed-down when ported to the GPU, and what kind of optimization techniques we should try to use to get higher performance.

Session 4 - Floating point

• Slides

In this video, we go through floating point issues, especially those related to parallel computing. This is an important topic because such floating point errors can determine the difference between a good and a bad result.