The Mont-Blanc approach towards Exascale
The Faculty of Informatics is pleased to announce a seminar given by Jean-François Méhaut and Brice Videau
DATE: Tuesday, February 19th, 2013
PLACE: USI Università della Svizzera italiana, room CC 321, Main building (Via G. Buffi 13)
The Mont-Blanc project (http://www.montblanc-project.eu/) aims to develop a European Exascale approach leveraging on commodity power-efficient embedded technologies. The project has developed a HPC system software stack on ARM processors, and will deploy the first integrated ARM-based HPC prototype by 2014, and is also working on a set of 11 scientific applications to be ported and tuned to the prototype system.
We present performance evaluation and analysis of two well-known HPC applications and benchmarks running on low-power embedded platforms. The performance to power consumption ratios are compared to classical x86 systems. Scalability studies have been conducted on the Mont-Blanc Tibidabo cluster. We have also investigated optimization opportunities and pitfalls induced by the use of these new platforms, and proposed optimization strategies based on auto-tuning.
Jean-François Méhaut is Professor of Computer Science at the Université Joseph Fourier (UJF) since 2003. His current research includes embedded systems as well as all aspects of high performance computing including runtime systems, multithreading and memory management in NUMA multiprocessors, multi-core and hybrid programming. He has participated in the European Exascale Software Initiative (EESI) as a member of the Software Eco-System group of experts. He is a member of the joint laboratory for petascale computing (JLPC) between INRIA and University of Illinois at Urbana Champaign. He is also PI of the FP7 IRSES HPC GA (High Performance Computing for Geophysics Applications).
Brice Videau is a CNRS Postdoctoral Fellow at Grenoble University since 2012. His current research includes embedded systems and high performance computing. He is participating in the Mont-Blanc European project, specifically on application porting, optimization and auto-tuning of HPC applications on low-power and hybrid high performance computing platforms. He is also actively involved in BigDFT, massively parallel electronic structure code based on wavelets, in which he developed an hybrid OpenCL implementation.
HOST: Prof. Rolf Krause and Dr. Achille Peternier