Guided research presentation. Ahmed is advised by Mario Wille.
Previous talks at the SCCS Colloquium
Ahmed Sami Fouad: Optimizing GPU Offloading with CUDA for a Patch-based Hyperbolic Finite Volume Solver in ExaHyPE
SCCS Colloquium |
The present study focuses on optimizing GPU offloading techniques within the Peano computational framework for adaptive mesh refinement (AMR) in Euler simulations using the ExaHyPE 2 engine. We first investigate the ExaHyPE 2 finite volumes Rusanov kernels to evolve the system of Euler equations using the CUDA framework. Then, we systematically identify and mitigate performance bottlenecks in the kernels' offloading process, yielding measurable overall improvements in the GPU computational efficiency.