Modern processors (both CPUs and GPUs) and architectures (grid, cluster, cloud) are particularly adapted for distribution, parallelism and vectorization (Single Instruction Multiple Data). Well-designed parallel code can run much faster than simple C/C++ implementations. In order to benefit from this speedup, one requires extensive knowledge of a variety of hardware and software technologies.
Our research and development activities in partnership with main technology vendors (Nvidia, Intel) allowed us to build expertise in HPC. This expertise helps our customers maximizing performance while focusing on algorithmic innovation (Automatic Differentiation e.g.).