Advanced LUMI course

This four-day course on-site in Amsterdam serves as a comprehensive discussion of the LUMI architecture and programming environment. It will include lessons on compiling and using software, programming models (HIP and OpenMP offload), porting, executing jobs, and optimizing applications to run on AMD MI250X. After completing the course, you will be able to work efficiently on both the CPU (LUMI-C) and the GPU partition (LUMI-G).

Please note that this is an advanced, comprehensive course with an emphasis on code development. LUST is also organizing recurring one- or two-day introduction courses to LUMI. These courses cover the most important topics needed to install and use software on LUMI, as well as run jobs efficiently on LUMI-C and -G. The next course will be announced shortly. All courses assume some familiarity with HPC, for example, from courses organized by most local compute centers, and are designed to focus on LUMI-specific elements.

Course Prerequisites

Participants are expected to have:

• Solid knowledge of HPC cluster computing.
• Familiarity with Linux command line and programming (C, Fortran, or Python).

This is not a high performance computing intro course but a in-depth discussion of the specifics and peculiarities of LUMI.

Learning Outcomes

After completing this course, participants will be able to:

• Understand the specifics and peculiarities of the LUMI hardware and software
• Write job scripts to get the best performance, for example through efficient GPU binding
• Compile programs utilize the Cray Programming Environment efficiently
• Profile their application in order to understand performance bottlenecks
• Use parallel debuggers to solve problems in MPI applications

Event info

The course will be jointly conducted by the LUMI HPE Centre of Excellence (HPE CoE), AMD and the LUMI User Support Team (LUST). Several staff members from HPE, AMD and LUST will be on-site, so this is the perfect moment for interaction with them.

The course is intended for users with ongoing projects on LUMI, users with project proposals in one of the national or EuroHPC channels, and support staff of local organizations of the LUMI consortium members.

The venue is at the offices of SURF in Amsterdam, the Netherlands. See the bottom of this page for more information on the venue. Participants are themselves responsible for all travel bookings.

If registrations exceed the number of planned participants, we will maintain a waiting list. The registration link can be found at the bottom of this page. The registration deadline is 24. October 2024, at 16:00 CET. As capacity is limited, participants who only intend to attend part of the course are kindly requested to leave their spot to others who wish to benefit from the entire course.

Attendance are free of charge.

To participate in the hands-on exercises and access the slides and training materials on the course day, participants will need to join a course-specific training project. This project is managed by Puhuri, which means that CSC users will need to set up a new LUMI user account. Step-by-step instructions will be sent out after registration on how to join the project and set up an account if needed.

Agenda

The main topics day by day. Each day 9:00 til 17:00 CET.

Day 1 – Mon 28.10. – Overview over LUMI hardware and Cray programming environment incl. compilers and parallel programming models.
Day 2 – Tue 29.10. – Debugging, introduction to AMD ROCm and HIP, Slurm binding, additional software on LUMI.
Day 3 – Wed 30.10. – Profiling CPU and GPU applications.
Day 4 – Thu 31.10. – I/O, advanced ROCm tools, LUMI support.

Registration

Register here: https://ssl.eventilla.com/event/kNLpb

Please register for the course by 24. October 2024, at 16:00 CEST. We reserve up to 10 in-person places for people traveling to Amsterdam from abroad, and they will receive a quick confirmation on a first-come, first-served basis so that they can arrange travel. Other participants will receive confirmation shortly after the deadline. If your plans change, we kindly ask you to cancel your registration as soon as possible, ideally before the registration deadline. The email acknowledging your registration will contain a link to manage it.

For those interested in the course who are unable to join in-person, please send us an email at jorn.dietze@uit.no, and we will find a solution.

Users who do not have an account on LUMI yet will receive temporary access for the purpose of the course. The allocated compute time should only be used for the course exercises. Any abuse will result in removal from the allocation for this and future courses.

Venue

The course will take place at SURF in Science Park 140, 1098 XG, Amsterdam.