The projects develops a novel algorithm for large-scale three-dimensional topology optimization that can run efficiently on both CPU and GPU architectures. The implementation is written in JAX and uses just-in-time compilation to generate high-performance numerical kernels while retaining the flexibility of a differentiable programming framework. The main objective is to evaluate the computational performance, scalability, and memory efficiency of the proposed method across increasingly large structural optimization problems. Problem sizes will range from small validation cases to approximately 128 million finite elements, enabling an assessment of performance at scales relevant to high-performance computing. The proposed method will be compared against established baseline implementations from the topology optimization literature. GPU performance will be benchmarked against a Futhark-based GPU implementation, while CPU performance will be compared against an efficient MATLAB-based implementation. The comparisons will focus on runtime, scalability, and memory requirements under comparable problem formulations. The project aims to demonstrate the potential of JIT-compiled differentiable programming for high-performance topology optimization and to establish a foundation for future extensions to more complex structural mechanics and optimization problems.