Computer Architectures for HPC

Duration

1 day

This course will usually run at least twice each academic year. If you would like a bespoke version of this course run in your department then please contact us.

Suitability

Research postgraduate students and above; teaching and lecturing staff.

Course content

This course is designed for HPC users who already write or develop codes to give an overview of the hardware and infrastructure code runs on and how design choices can effect how efficiently code runs.

The course starts from first principles.

At the end of this course, participants will:

  • Understand the structure of a typical single core and multicore CPU
  • Understand the building blocks of a typical HPC cluster
  • Have an overview of how compilers create executable code from source code
  • Understand the concepts of memory bandwidth and NUMA (non-uniform memory architecture)
  • Understand the differences between the different filesystems on the HPC service and how they are most effectively used

Prerequisites

The course is designed for those with some familiarity of the UNIX environment and experience of programming. No previous experience of MPI or parallel programming is necessary. Experience of compiling and submitting jobs on the ARC facilities is useful.

Booking

This course can be booked through the IT Training Unit