Many scientific problems are of such complexity that solutions are obtained only by using a wide variety of computing hardware ? all at once. The concurrent use of e.g. multiple clusters, Grids, and Clouds, has spawned a new computing paradigm: ?Jungle Computing? *1+. The Ibis eScience software framework [2,3] enables easy implementation and deployment of large-scale Jungle Computing applications consisting of any variety of tasks implemented using any popular language or model (e.g. C, MPI, Python, CUDA). It also can select one of several alternative implementations (?equi-kernels?) that calculate the same result using different hardware. This project aims to apply Ibis, with the requested hardware infrastructure, to make significant steps forward in solving two important scientific (multi-model / multi-kernel) problems: 1. Star Cluster Evolution (Leiden) The early evolution of star clusters in the galaxy is a multi-physics problem, with spatial scales covering 14 orders of magnitude between the smallest relevant structures and the size of a cluster. The range of physical domains (gravitational dynamics, stellar evolution, radiative transport, hydro-dynamics) renders modeling the birth of star clusters one of the biggest challenges in computational astrophysics. Although the physical domains are all modeled effectively using the AMUSE framework (comprising multiple equi-kernels) , gravitational dynamics is best solved using a GPU, while stellar evolution requires a cluster (or Grid, cloud), and radiative transport is best conducted on a supercomputer. Full simulations must therefore be performed using the Jungle Computing paradigm. 2. Climate Simulations (Utrecht) Climate models are essential tools to investigate the effects of atmospheric greenhouse gases on climate changes. Such models are composed of ocean, atmosphere, land, ice and biosphere models, which are integrated by a so-called ?coupler?. Today, climate simulations are performed on a single supercomputer and the coupler is limiting the scaling of the full model. As some of the model components have a better performance on a GPU (e.g. the ice model, the ocean model (i.e. the Poisson solver)), it is beneficial to run different model components on different platforms. We aim to perform full climate model simulations (with the Community Earth System Model ) using the Jungle Computing paradigm.