From October 13 to 17, the 7th MareNostrum Hackathon took place at the Barcelona Supercomputing Center (BSC) headquarters, Spain. This annual event, organized by BSC, once again brought together HPC developers and researchers with the aim of helping participants to improve their applications with the guidance of expert mentors.
This year, the hackathon introduced two specialized, non-simultaneous tracks:
- Track 1 (MN5): October 13–15, focused on Europe’s pre-exascale supercomputer MareNostrum 5.
- Track 2 (RISC-V): October 16–17, dedicated to exploring open hardware architectures.
Our group member Xavier Sáez participated in the MN5 track to continue advancing the porting of the plasma fusion code BIT1 to GPU architectures. The focus was on improving the efficiency of data movement between the CPU and GPU.
During the event, he used Nvidia Nsight to visualize and analyze the performance of BIT1 on a GPU with the help of Filippo Spiga (NVIDIA). This analysis enabled him to identify unnecessary data transfers and optimize the communication between the CPU and GPU. Additionally, he worked on maximizing the number of routines ported to the GPU to minimize CPU-GPU data exchanges and enhance overall code performance.
The event also featured two technical talks: “Unified Virtual Memory and NVIDIA Nsight System” by Fillipo Spiga (NVIDIA), and “Open OnDemand: Democratizing Access to HPC Resources” by Lorenzo D’Amore (Do IT Now).
MNHACK25 once again proved to be an excellent opportunity to exchange ideas with experts and further improve the efficiency and scalability of fusion-related computational codes. The Fusion Group would like to thank the MNHACK25 organizing team for this valuable and inspiring event.