Orateur
Description
Summary
Computing industry still continuously improves processors performances, although it is not any more thanks to a frequency rise. The trend is now to increase the number of cores. This is cheaper for what concerns power consumption, yet those new cores are lighter than traditional processors, and individually benefit from a smaller memory. In this future many-core era, the huge legacy code of high energy physics could turn completely unusable, unless we deeply refactor it so to embrace massive parallelism.
This aim of GridCL project is to study the transformation of P2IO software applications in order to exploit efficiently today what looks like most the future many-core hardware : the graphics coprocessors from NVIDIA and/or AMD, and the new "Many Integrated Core" architecture from Intel. We plan the integration of such hardware in computing grids, which are inherently heterogeneous. This is why we favor portable software solutions such as OpenCL, rather than specific dedicated products such as CUDA or TBB.