NVIDIA GPUS POWER LATEST SUPERCOMPUTING AND HIGH PERFORMANCE COMPUTING ADVANCES

Thomas Schulthess is the director of the Swiss National Supercomputing Centre (CSCS) at Manno. He studied physics and earned his Ph.D. degree at ETH Zurich. As CSCS director, he will also be professor of computational physics at ETH. He worked for twelve years at the Oak Ridge National Laboratory (ORNL) in Tennessee, a leading supercomputing and research in the US. Since 2002, he led “Computational Materials Science Group” with 30 co-workers.

My experience is focused on high performance (parallel)-scientific computing applied to numerous problems of research and industrial interest. I have a strong interest in parallel programing and computing. I had the opportunity to work in several research centers/universities, such as the University of Manchester, BCAM, CIEMAT, UCM, UCLM, among others, and projects (NLAFET, FeniCS-HPC, Alztools, SyeC, SCORE), where I worked on multiple areas such as, Linear Algebra, Computational Fluid Dynamics (CFD), Lattice-Boltzmann Method (LBM-HPC tool), Partial Differential Equations, Finite Element Method (FeniCS-HPC tool), Fluid-Solid Interaction (Lattice-Boltzmann Method and Immersed Boundary Method), Mesh Refinement over LBM, Non-uniform geometries applied to LBM, Many-Task Computing on Heterogeneous Architectures (NVIDIA GPUs), Brain-Biomedical and Satellite Image processing (ITK, OTB, SPM packages), Distributed Database Systems (metric spaces and similarity search), Formal Methods (ROSA Algebra) and Software Engineering (Apadyt tool). The use of heterogeneous multicore-manycore (NVIDIA GPUs) systems has been and continue being an important part of my work. To enhance the efficiency of these platforms, I have developed multiple heterogeneous schedulers, in which the computational load is distributed statically either to hardware accelerator or multicore, depending on the computational cost and intrinsic characteristics of those problems to be computed. Recently, I was involved, under the supervision of Prof. Jack Dongarra, in the development of a new BLAS (Basic Linear Algebra Subroutines) standard to compute efficiently multiple “small” BLAS problems over current multi/many-core architectures (Batched BLAS), as well as the developing of the new release of the PLASMA library using task-dependent OpenMP. Previously, my work consisted of implementing a new hybrid (OpenMP-MPI) open-source tool for simulating incompressible flows based on Lattice-Boltzmann Equation (LBM-HPC) over supercomputing systems. One of my current research lines consists of the study of new grid geometries and grid refinement techniques imposed by error control applied to the Lattice-Boltzmann Method for improving its numerical accuracy, scalability, and performance. In my current position at Barcelona Supercomputing Center (BSC), I work in the analysis and implementation of numerical methods making use of new features that OmpSs, CUDA, OpenACC provide. Also, I am part of the Human Brain Project team and NVIDIA Center of Excellence team at BSC. In particular I am in charge of the acceleration of the simulation of the behavior Human Brain using NVIDIA GPUs. I have authored over 30 publications, including 10 journal citation report (JCR) papers and about 30 peer-reviewed international conference papers.