Introduction to GPU computing on HPC: Glossary

Key Points

Intro to GPU computing
  • GPU vs CPU

  • Suitable problems for GPU

  • CUDA, and high level GPU environments

  • Where to get one

Writing and running your first GPU program
  • Compilation of C and CUDA code are siilar

  • Running serial code and GPU code is similar

Connecting to Artemis HPC
  • Connecting to Artemis requires a terminal emulator, and an account with access.

  • Users connect to Artemis’ login nodes only.

  • On Windows, use WSL, PuTTY, or another shell and terminal application of your choice.

  • GUI login access is also available.

Running on HPC
  • Copy relevant files to directories.

  • Load modules and compile code.

  • Submit jobs to the scheduler.

Matlab GPU example
  • You can optimise your code for GPU in different ways

  • Matlab has inbuilt functionality to run on GPUs

  • Submitting a Matlab job is otherwise the same as submitting any job on HPC

Deep Learning Time Series with Python, tensorflow, and a GPU
  • Artemis has lots of software, but sometime you need to roll-your-own

  • Getting the correct balance of versions for software stacks is imperative

  • Not all GPUs are compliant with all software