Given a design problem, the topology optimization (TO) method can be used to programmatically find an optimal solution. At its core, TO relies on an iterative approach where, at each iteration, the design shape is modified based on simulation data. The time to compute a solution strongly depends on the performance of the simulations. While multi-GPU systems are effective at accelerating simulations, achieving the best performance and scalability has its challenges. Along with domain expertise, a strong knowledge of parallel programming is required. We'll present Neon - a new framework designed to make multi-GPU programming easier and more intuitive for non-GPU experts. Neon is based on a structured parallel model that primarily targets simulation on cartesian grids. Neon efficiently hides the complexity of managing a domain that is partitioned across multi-GPU and more. We'll showcase preliminary performance on structural and fluid simulations.