Best Practices in Running Collaborative GPU Hackathons: Advancing Scientific Applications with a Sustained Impact


Best Practices in Running Collaborative GPU Hackathons: Advancing Scientific Applications with a Sustained Impact

Chandrasekaran, S.; Juckeland, G.; Lin, M.; Otten, M.; Pleiter, D.; Stone, J. E.; Lucio-Vega, J.; Zingale, M.; Foertter, F.

Abstract

The OLCF GPU Hackathons are a one-week code-development/learning event to better enable attendees to utilize GPUs. It only took three years to grow from a ``Let's give this a try''-event to a repeatedly copied format with several spin-offs that inspired HPC centers around the world.
Sticking to a few fundamental principles---work on your own code, learn from your mentors just what you need and when you need it, stay flexible in achieving your goal---the week-long hackathon format created at Oak Ridge Leadership Computing Facility (OLCF) has been just the spark needed by many groups of scientists to light the fire of a wider GPU adoption in leading-edge as well as university-scale HPC environments. Most interestingly, the format enabled both ends of the experience spectrum---graduate students vs. postdoc fellows---the same kind of progress and chance of success.

Keywords: B.1.4.b Languages and compilers; D.1 Programming Techniques; M.12.0.b Case Studies in Scientific Applications

Downloads

Permalink: https://www.hzdr.de/publications/Publ-27612