# HPE/AMD/PDC/ENCCS Dardel hackathon March 7 & 14-16 Todo: 1. Decide on questions/information for submitting projects 2. Mentors 3. Event page text 1. Submitting projects - code name - domain - what does the code do - programming language and libraries - GPU programming framework you intend to use - brief description of algorithm(s) - current performance, bottlenecks, profiling results - what would be your goal at the hackathon? 2. Mentors - 4 from AMD - 2 from HPE - 4 from Sweden - recommended hotel - need to decide budget with PDC - at least 2 preferably 3 people per project - diversity goal "we particularly recommend underrepresented people to apply" - HPE needs to send onsite people - 1 onsite mentor per team, another on-call who can come in and work with teams - morning debrief every day - template slides: progress made, challenges, requirements/needs - use Slack channel for communication - social events: catering, dinner, drinks, get-together evening before hackathon - George and Bob will come - at least one HPE person onsite, be able to pull in compiler expert etc - send doodle for meeting later in december (12-16/12?) - provide profiling instructions so teams can show profiling results on day 1 march 7 evaluation criteria: - skills needed by mentor, availability of expertise - diversity in project team - the code should be used by more than one research group - open source requirement - potential for impact for a wider community - at least 3 project members - should have a scale or efficiency goal - should have a build system that can be made to work on Dardel, and will be prepared to compile the code before hackathon - be prepared to run a profile before hackathon (high level, where are bottlenecks etc) - Scalasca - Score-P - arm-forge # Event page ## HPE/AMD/PDC/ENCCS Dardel Hackathon ### About the event Dardel is a Swedish flagship supercomputer operated by PDC at KTH in Stockholm. It is an HPE Cray EX system with AMD EPYC 64 core CPUs and AMD Instinct MI250X GPUs. Compute time on Dardel is available for academic research via SNIC and PRACE as well as for business and industrial research projects. This Dardel hackathon is for teams of research software developers who want to accomplish one or more of the following goals: • Port their code to run on AMD GPUs using any of the GPU programming frameworks listed below • Improve the performance of code that already runs on AMD GPUs • Scale up their codes to multiple AMD GPUs During the hackathon, project teams will be paired up with expert mentors from HPE, AMD, PDC and ENCCS who will guide them during the hackathon to accomplish their stated goals. Teams should have at least 3 members who are prepared to work full time on their projects for the duration of the hackathon. We furthermore also strongly encourage diversity in project teams. ### Location The Dardel hackathon will take place over four days split over two weeks. The first day of the hackathon, March 7, will be online while the main work will be hosted on-site on KTH Campus in Stockholm, Sweden, on March 14-16. Further details about the location will be provided later. ### Requirements • Teams should consist of at least two, and preferably three, members who are ready to actively participate during the event. • Teams should know their application well and have the programming experience to modify the code. • Projects should be under an open-source license. No advanced GPU skills are required, but it is helpful for teams to know the basics of GPU programming and profiling. Team members should get familiar with aspects of ROCm and HIP (https://docs.amd.com/) ahead of the hackathon which are relevant to the project goals. Appropriate learning resources will be shared with accepted teams. ### Travel and accommodation Information for travel arrangements will follow. ### Tentative agenda March 7, 9:00-16:00 CET: 9:00-10:00 Introduction and practical information 10:00-12:00 Meetings between teams and mentors 13:00-16:00 Tutorials on the HPE environment and the AMD ROCm software ecosystem. March 14, 9:00-17:00 CET 9:00-9:30: Scrum session 9:30-16:00: Project work 16:00-17:00: Project summary and discussion March 15, 9:00-17:00 CET 9:00-9:30: Scrum session 9:30-16:00: Project work 16:00-17:00: Project summary and discussion March 16, 9:00-17:00 CET 9:00-9:30: Scrum session 9:30-15:00: Project work 15:00-17:00: Final project presentations ### Compute Resource Teams attending the event will be given access to Dardel ahead of the hackathon. ### Project submission To apply for participation in the Dardel GPU hackathon please describe your project in detail at WRITEME. Note that all team members should register and fill the “Personal information” section, but only one person in your team needs to fill the “Project description” section. Applications will be reviewed by an expert panel according to the following requirements: • Project team should consist of at least 2 members who know the code well. • Diversity in the project team. • The project should be described in sufficient detail and have well-defined goals for efficiency or for scaling up. • The code should have a significant impact (or large potential for significant impact) for a specific organisation or a wide community • The code should be open source. A link to the source code should be included in the application. • Projects should have a build system (e.g., Make or CMake) which can be made to work on Dardel, and teams should be prepared to compile the code on Dardel before hackathon. • Availability of expertise needed to successfully support the project. Ahead of the hackathon, accepted teams should be prepared to: • Go through learning resources on the AMD ROCm software ecosystem recommended by the organisers. • Profile their code ahead of the hackathon to identify performance bottlenecks using tools like Scalasca, Score-P or Arm Forge (instructions will be provided if needed). ### Registration (This will be moved into the registration form) Please fill in this registration form: LINK Note that all team members should register and fill in their personal information, but only one member should fill in the “Project description” section. Personal information: - First name - Last name - Phone number - Affiliation - Email - Country - Who else is in your team? - Experience on working with this application - Have you ever used AMD GPUs before? Project information: - What is the name of the code? - What is the name of your team? - Please list all team members along with their email addresses and (if available) GitHub profiles. - Scientific domain - What does the code do? - Please provide a link to the code repository. Alternatively, you can send code by email to thor.wikfeldt@enccs.se. - Which programming language(s) does the code use? Which libraries does it depend on? - How do you compile your code? - Please describe which GPU programming framework you want to implement. - Please provide a brief description of key algorithm(s) in the code. - Please describe the code’s current performance, bottlenecks, and any profiling results you might have. - What is your team’s goal at the hackathon?