Intel® Parallel Computing Center at CERN, European Organisation for Nuclear Research

Principal Investigators:

Federico Carminati works at CERN, (Geneva Switzerland) where he leads the detector simulation activities. After his degree in 1981 he worked at Los Alamos and Caltech before coming to CERN, where he was responsible for the CERN Program Library, the worldwide standard High Energy Physics code in the 80’s and 90’s. From 1994 to 1998 he worked with Nobel Prize Carlo Rubbia at the design of a novel accelerator-driven nuclear power device. From 1998 to 2012 he was Computing Coordinator of the ALICE experiment at LHC. In 2013 he obtained his PhD in physics from the University of Nantes.


The ever-increasing need for computing resources has prompted a sustained effort to optimise the High Energy Physics program library to take advantage from the new computing architectures and accelerators. This is necessary if we want the HEP code to continue to profit from the nominal increase in performance of the new computing devices following Moore’s law. After a detailed R&D phase during 2012 and 2013, CERN decided to target simulation as a primary candidate for optimisation. Half of the 300,000 cores constituting the World LHC Computing Grid are continuously running simulation programmes, and the simulation needs are expected to increase at least tenfold in the near future. The studies conducted by a team of experts within the SFT group concluded that to profit from the new hardware capabilities, the existing simulation code (Geant4) had to be substantially rewritten. A project was launched to develop the new generation of detector simulation framework with improved physics and optimised performance. This project, called GeantV, has developed a parallel simulation framework and highly optimised libraries for geometry. The new geometrical components achieve speedups that in some cases exceed one order of magnitude, and will be offered as an alternative also for the Geant4 system. By the end of the current year the GeantV prototype aims to simulate the full CMS detector, initially with simplified physics and sub-optimal performance. We expect to have meaningful performance figures for such a realistic application during 1Q 2015, the goal being to have a performance improvement of a factor between 3 and 5 with respect to current simulation codes, with a long term objective to obtain one of order of magnitude improvement in performance.

The work covered by this Intel® Parallel Computing Center (Intel® PCC) within the GeantV project aims at providing, in the first year, the first GEANT-V version that is vectorisable and thread wise scalable on Intel® Architectures demonstrating a speedup of a factor between 5x and 10x over the scalar version on a simplified example. Target architectures will be Xeon PHI (Knight Landing) and Xeon (Haswell). A tech paper will be presented at the Computing for High Energy Physics conference 2015. The second year will be dedicated to the fine targeting further performance demonstrating a speedup of a factor between 2 and 5 on a complex example. A new assessment will be made whether a speedup of a factor 10 is attainable and how. Documentation and benchmarks results will be published on applied optimization technics. A public Technical paper addressing GEANT-V modelling will be presented (CHEP 2016), as well as a joint paper CERN- Intel promoting GEANT-V.


Related Websites:

For more complete information about compiler optimizations, see our Optimization Notice.