Prev: P12.4 Next: P12.6

P12.5: Goz, David
David Goz (INAF - OATs)
Sara Bertocco (INAF - OATs)
Luca Tornatore (INAF - OATs)
Giuliano Taffoni (INAF - OATs)




Theme: Algorithms
Title: Astrophysical codes migration into Exascale Era

The ExaNeSt H2020 EU-funded project aims to design and develop an exascale ready prototype based on low-energy-consumption, ARM64 cores and FPGA accelerators. We participate to the application-driven design of the platform and to the validation of the prototype. To carry on this work we are using three widely used astrophysical codes: a direct N-body code, called Hy-Nbody and derived from HiGPUs (Capuzzo-Dolcetta+13), to simulate stellar cluster dynamics; PINOCCHIO (Monaco+02), a code aimed to generate catalogues of dark matter halos and their merger histories; GADGET-3, evolution of the public code GADGET-2 (Springel+05), a state-of-the-art N-body and hydrodynamical code for large-scale, high-resolution numerical simulations of cosmic structure formation and evolution. Core algorithms of the aforementioned applications have been improved in such a way to increasingly fit to the exa-scale target platform. Waiting for the ExaNest prototype release, we are performing some tests and code tuning operations on ARM SoC facility. We deployed a testbed HPC cluster based on 64-bit ARMv8 Cortex-A72/Cortex-A53 core design, powered by the Mali-T860 embedded GPU, and SLURM Resource Manager. In parallel we are porting a kernel of Hy-Nbody on FPGA aiming to obtain high performance-per-watt solutions for our algorithms. In this poster we describe how we re-engineered the applications and we show first test results on ARM SoC.

Link to PDF (may not be available yet): P12-5.pdf