Experiences of porting real codes to Intel's Xeon Phi

IntelŐs Xeon Phi is a many-core co-processor which currently powers the fastest supercomputer in the world, Tianhe-2, offering around 1 TFLOP/s in a PCI package, similar to modern GPUs. As with any new architecture, applications must be ported to take advantage of the performance offered by the hardware. This talk will summarise the hardware and software architecture of the Xeon Phi, and discuss some of the challenges encountered and lessons learned while porting CP2K, work carried out in 2013 under the PRACE project. We will also briefly discuss the next generation on Xeon Phi, and how it may help to overcome some of these issues.