The hardware characterstics of Graphical Processing Units (GPUs) have advanced to the point that two order of magnitude performance improvements can be obtained in problems that are of general utility.
I will outline the design of the code GLISSE (GPU Longterm Integrator in Solar System Evolution), illustrating how careful choices are required to maximize the computational power of the GPU hardware.
I will then describe several tests to illustrate scientific exploration of the stability of orbits (1) between Uranus and Neptune, and (2) in the Kuiper Belt beyond Neptune. For these applications GLISSE is 300-1000 times faster than a CPU core, and thus the hardware cost makes it vastly more financially effective to use GPU hardware rather than clusters of CPU nodes.
We have extended the integrator's capabilities to handle small bodies
encountering planets, handling such relatively rare encounters on CPU cores while the GPU deals with huge numbers of test particles.
I will discuss the application of this GLISSER code (the R standing for 'regularized') to the orbital evolution of the Kuiper Belt region under the existence of additional planets in the early Solar System.
Francesco Marzari
