Introduction to GPU computing on HPC

Synopsis: Introduces GPU computing, and running GPU jobs on Artemis and other HPC systmes. The University of Sydney’s Artemis HPC hosts several NVIDIA V100 GPUs. This course will help you to understand basic concepts of GPU programming. You will learn fundamentals of basic CUDA code, and write and run examples using C/CUDA, Matlab, and Python. You will undertake practical applications in Deep Learning using Python, Tensorflow, and Keras. You will learn how to set up suitable environemnts on Artemis for GPU-enabled applications to run, and how to run and submit jobs on the Artemis HPC GPU queue.

Target audience: Students and staff who would like to learn how to use GPU enabled code, especially for Artemis HPC. Participants must have a valid USyd unikey.

Follow-on courses: This course is part of the Artemis HPC Training Series.

This first lesson is intended as part one of a pair, with Introduction to the Artemis HPC in the morning and Data transfer and RDS for HPC in the afternoon. It is recommended to register for both, however you may choose to take these courses on separate days as suits your needs.

Another course in the series, Intermediate Artemis HPC, extends users’ knowledge of the PBS Pro scheduling system, and builds scripting skills for automating large workflows on HPC.

A specialised lesson, Matlab on Artemis: The MDCS, is a course for users of MATLAB who would like to submit jobs directly to the cluster from their local computers.

Everything covered here has more detail in the Artemis guide, get it here.

Prerequisites

Competency on the Unix/Linux command line:

  • If you are interested in learning HPC but have no Unix/Linux command-line skills, you MUST first take an Introduction to Unix/Linux course.

OWN LAPTOP REQUIRED.

Schedule

Setup Install software and data needed to begin the lesson
00:00 1. Intro to GPU computing
00:30 2. Writing and running your first GPU program
00:30 3. Connecting to Artemis
00:30 4. Matlab GPU example
00:30 5. Deep Learning Time Series with Python, tensorflow, and a GPU
00:30 Finish

The actual schedule may vary slightly depending on the topics and exercises chosen by the instructor.