Hackathon: Optimizing for AMD GPUs
The Hackathon is a way for you to get free help to optimize your software for the AMD GPUs available on LUMI-G!
You can apply if you are a researcher or a company located in either one of the LUMI consortium countries or a member of EuroHPC country. Ph.D. students and holders of postdoctoral fellowships are also eligible to apply. Up to six (6) projects will be selected for this hackathon. The selected projects will receive help from HPE and AMD staff of the HPE Center of Excellence for LUMI and from members of the LUMI User Support Team. The Hackathon will be on site at CSC in Espoo (close to Helsinki). The deadline for applications is February 24 2023.
There will be another hackathon in the second half of 2023.
The following sections give more detail on what to expect and describe how you can apply. In order to make sure we can properly support those who attend we are asking for some information on what you hope to achieve.
Who can apply
This is an open call. Anyone who is a member of an organization eligible to receive an allocation for LUMI compute resources is also welcome to apply to this program. In brief, this means that a research software developer affiliated with a research organization in the EuroHPC countries and/or countries part of the LUMI consortium can apply. A company, which has its headquarters in such countries, is also eligible. By partnering with European research groups and/or companies, a non-European applicant can also participate, but please note that the principal investigator of any projects that apply will need to be based in the EU or an associated country. Ph.D. students and holders of postdoctoral fellowships are also eligible to apply, but we may ask for further letters of support in such cases.
For more information, please see:
- EuroHPC JU Access Policy https://eurohpc-ju.europa.eu/sites/default/files/2021-03/Decision%2006.2021%20-%20Access%20policy.pdf
What we are looking for
We are looking for research groups and companies interested in making their scientific software utilize efficiently the new AMD GPU-compute architecture in LUMI, in collaboration with LUMI experts. We are looking for proposals outlining a team effort of solving performance bottlenecks in existing software. The scope should be limited so that the task can be accomplished in one week with your team of two (2) to four (4) people.
Collectively the team should know the application intimately. There will be intensive mentoring during this 5-day hands-on workshop. Mentors come from HPE’s Center of Excellence, and the LUMI User Support Team and will help your team accomplish your team’s goal. We expect that your software already compiles and runs on AMD GPUs, preferably with a build system based on standard tools like Make or CMake.
It is an advantage if intended work is mainly in their own source code, and not in external third-party libraries (for which they are not the main developers), as it may be difficult to interact quickly with third parties during the hackathon.
We are mainly looking for software with open source licenses, or at least free academic use. We will consider commercially licensed software only in exceptional cases.
How to apply
Closing date: 24 February 2023, 17:00 CET
Hackathon period: 17-21 April 2023
We will organise an accompanying course about profiling GPU programs the week before the hackathon. It will happen online on the 13 April from 9:00 until 17:00. Attendance from teams joining the hackathon is advised.
Please register here: https://ssl.eventilla.com/event/yWdEZ
The application form is relatively short and we ask for: contact information, a short technical description of the software (programming languages, libraries etc), a description of the intended work, and a motivation for why a successful effort will useful for the end-users (scientists).
Teams whose application successfully passes the selection process will need to register for the event to confirm participation within two weeks from notification of acceptance. The notification of acceptance will include information on how to register.
How we will assess your application
The teams will be selected based on a scientific and technical review. The scientific review is performed by the LUMI consortium and/or external evaluators. The technical review is performed by the LUMI User Support Team and HPE/AMD Center of Excellence (“CoE”). We may ask for an interview to discuss your proposal (done remotely). In such case, applicants will be provided with comments prior to the interview in order to prepare. The interview will only be focused on the technical feasibility and not on the scientific merit.
The scientific assessment is mainly done with respect to the potential impact in the specific field of science where the software will be used. Relatively less importance will be paid to the scientific track record of the applicants. Furthermore, the proposed software and/or implementation is evaluated relative to the state-of-the-art in the field, and if a successful effort would fill a gap in the currently available software on LUMI.
The technical assessment is mainly done with respect to the technical feasibility of porting the software to AMD GPUs and subsequently running on the LUMI supercomputer. These include both: 1) technical aspects of the software such as programming language, the use of libraries and frameworks, and the internal data structures and algorithms, and 2) the practical aspects such the difficulty of achieving the goal during the hackathon and the team’s ability to participate on-site. In particular, the software development track record of the applicant (or applicants), and the balance of skills in the final project team, which will include staff from the HPE CoE and the LUMI User Support Team, will be taken into account. Projected software speed-ups or preliminary benchmarks are helpful in the evaluation, but are not required.
The hackathon will be held at CSC – IT center for science in Espoo, Finland on April 17-21. It will start around lunchtime on the 17th and end at lunchtime on the 21st to allow for travel. The region is served by Helsinki-Vantaan airport. It takes approximately 1h to reach CSC from the airport. We do not provide accommodation.
For any other specific information, please contact:
- LUMI User Support
- Emmanuel Ory, LUMI User Support Team leader, emmanuel.ory(at)csc.fi
- Detailed technical information about LUMI can be found in this blog post.