Frames per Second Laser Plasma Simulations - Making large scale simulations really fast ... or slow

Many-core compute architectures such as graphic cards will be the building blocks of next-generation Exaflop computers. With these architectures, complex laser plasma simulations can run on a frames-per-second rate, decreasing the waiting time to get results to hours instead of weeks. Thus, large surveys for optimum acceleration parameters come in reach. This will enable theoreticians and experimentalists to discuss and understand the physics behind particle acceleration scenarios instead of simply adding a few pretty simulation pictures to a publication. When considering future applications of ultra-intense lasers, new physics will have to be taken into account. It is thus not only mandatory to make simulations fast, but to then add new physical effects into the code. This will require new strategies to leverage the power of next-generation supercomputers. We propose potentially successful techniques to get the most out of upcoming HPC systems based on our experience with PIConGPU and show what is already possible today.