Speeding-up image processing with multi-core CPUs and GPUs

Benjamin Ranft - FZI Research Center for Information Technology Karlsruhe, Germany
Multimedia Communications

Date: -
Location: Eurecom

The computing power of modern processors can only be utilized by parallel software. This presentation gives a general introduction to programming multi-core CPUs and their SIMD units, as well as graphics processing units (GPUs). It also illustrates how performance and energy efficiency can be optimized on a heterogeneous system, which combines both CPUs and GPUs. Achievable speed-ups are presented using dense stereo vision as a sample application.