Graphics Processing Units (GPUs) offer high performance and massive parallelization, but learning how to program GPUs for scientific applications can be daunting. To help reduce the barrier to entry, Princeton will host its 2nd annual GPU hackathon, organized and sponsored by NVIDIA, Oak Ridge Leadeship Computing Facility (OLCF), and Princeton Institute for Computational Science & Engineering (PICSciE)and OIT Research Computing.
Interested research groups should apply to send a small team of 3-6 developers who will be paired with experienced GPU mentors from industry and various national laboratories in order to help migrate their code to GPUs and/or optimize codes already running on GPUs, all during an intensive 5-day scrum. Prior experience with programming GPUs is not required, but participants will be encouraged to get an overview of different GPU programming paradigms prior to the hackathon via training materials that will be provided in advance.
A guide for what to expect as an attendee can be found here. The inaugural 2019 GPU Hackathon at Princeton followed this format and was a great success. All participating teams achieved a code speedup of at least a factor of 5x, with some teams gaining the equivalent performance of ~200 CPU cores for their codes from just a single GPU.
All fields are welcome, from astrophysics to machine learning to genomics. Although Princeton and Princeton-area teams will be given priority, a Princeton affiliation is not required, so please forward this announcement to interested colleagues from neighboring institutions (partner GPU hackathons hosted in other parts of the US and the world are listed at gpuhackathons.org/events). Prior hackathon participants are also welcome to apply, but they should explicitly address in their applications the additional value they hope to gain from attending the hackathon again.
Please note that acceptance at this event is *not* automatic. Space is limited, and applications will be reviewed for goodness of fit with the goals and resources of the hackathon (including the availability of appropriate mentors for a given team). The (brief) application can be found at the main GPU Hackathons application website. APPLICATIONS MUST BE RECEIVED BY NOON (EST) ON MONDAY, MARCH 2, 2020 for full consideration.
Access to computing resources will be provided for the duration of the hackathon, along with lunch, coffee/snacks, and a workshop dinner on one of the nights. Once final teams are selected (by early April), a landing page will be announced with information on these details as well as with travel and lodging options for participants from outside Princeton.
Finally, if you already have some GPU programming experience but do not plan to submit an application, and if you would like the opportunity to spend a week working with and learning from a bevy of GPU experts, we will be looking for a handful of "junior mentors" from the Princeton area to pair up with teams at the hackathon. If you are interested in being a junior mentor, or if you have any other questions, please contact Gabe Perez-Giz directly.