The Hackathon consists of two parts
-
a preparation phase with a discussion of your project and goals for the hackathon, taking place November 20 (online),
-
the actual coding phase, where you will work towards the defined goal, see below, and port your code to Hunter, taking place December 1-5 (hybrid).
HLRS's next supercomputer system called “Hunter” is available. As for every new system, users have to spend some effort into porting their code and workflow to the new environment. Furthermore, the vast majority of Hunter’s compute power will be delivered by the GPU part of the APU, which requires users to adapt their hot loops in order to offload them to the GPUs. In this hackathon we will hence support our users in doing both. This hackathon focuses on scientific codes but also dedicates time to users running part or all of their workflows with AI.
In order to offload to GPUs, multiple programming models (HIP, OpenMP device offloading, PSTL/do concurrent) are available depending on the programming language used. If you are unfamiliar with them, we offer a recap of OpenMP target offloading in the course Introduction to OpenMP Offloading with AMD GPUs. In the preparation phase, pros and cons of those models based on your situation could be discussed if you are just starting porting, and you and the support staff will decide together which are suitable goals for the coding phase and how to proceed.