AMD-GPU Workshop, Lectures and Exercises

Start: Apr 21, 2026 09:00 AM (Local Time Germany)
End: Apr 24, 2026 01:00 PM
Speaker: AMD
Location: online via Zoom
Host: MPCDF
Contact: training@mpcdf.mpg.de

This is the first part of a two-part workshop together with AMD to teach the usage of the MI300A GPU which is employed in our VIPER-GPU supercomputer. On four half-days, experts from AMD will present different topics related to this new APU architecture. Together with exercises on the AMD cloud system, you will learn about HiP and the transfer of a CUDA code to HiP, about using OpenMP TARGET directives for GPU porting in a portable way, standard language constructs and the toolsuite that comes with the ROCM system.The second part will follow the week after with a hackathon where you can bring-in your own codes and we will work on them together to reach porting or performance goals on our VIPER-GPU system. Please register for the two events separately.

In this workshop, experts from AMD and Eviden will focus on the how to port a code from NVIDIA GPUs to MI300A. Knowledge of GPU programming in general is a prerequisite.

This lecture part of the AMD-GPU training is held together with the HLRS in Stuttgart, which also has a MI300A installation. The detailed program can be found under https://www.hlrs.de/training/2026/gpu-amd.

The registration for this event is handled by HLRS. So please go to https://regi.hlrs.de/2026/GPU-AMD/registration/ for registration. Use your work email and MPG affiliation in the registration form. For the exercises in this lecture part, you will get access to the AMD cloud system AAC to unify the user experience during the workshop. But you can also try all exercises on our VIPER-GPU system, the hardware is the same and differences in the software stack will be mentioned in the presentations.