Dear users,
on May 22th at 11 am at CINECA (Casalecchio di Reno, Bologna) , Dr. Luca Franci by University of Parma will present the seminar titled:
"Possible optimization and porting of a numerical code for 3D simulations in General Relativistic Magnetohydrodynamics on CPU/GPU hybrid clusters"
Abstract:
"General Relativistic Magnetohydrodynamics (GRMHD) is the study of relativistic magnetized flows in very strong gravitational fields and it’s therefore the right framework for modeling compact objects like black holes and neutron stars, which are believed to be responsible for many high-energy phenomena in astrophysics.
X-ECHO (Del Zanna et al. 2007, A&A, 473, 1; Bucciantini et al. 2011, A&A, 528, A101) is a numerical code aimed at performing GRMHD simulations, such as the 2D evolution of isolated magnetized neutron stars. The GRMHD conservation laws are here solved within a finite-difference discretization scheme whereas a staggered constrained-transport method is used to preserve the divergence-free condition for the magnetic field.
Since all the computational operations are local, the problem is naturally data-parallel via domain decomposition. Actually, X-ECHO is only MPI-parallelized along one direction of the full 3D domain and the parallelization is not optimized at all, so that the code can be run only on a very few processors (up to 8) and one is forced to strongly reduce the size of the grid and/or the evolution time in order to get results in a reasonable time. A way to get X-ECHO run faster is then to improve the MPI-parallelization and to port it on GPUs.
There’s already a GPU-accelerated code for solving the equations of GRMHD, which is the HORIZON code by Zink (arXiv:1102.5202v1): a simulation performed with a GPU can be at least 10 times faster than one performed on a quad-core CPU. Given the high computational cost of 3D full GR simulations, an order of magnitude in performance should be enough of an incentive to consider the porting of X-ECHO on GPUs."