Together with colleagues from Umeå University, we implemented in Nek5000 the possibility to perform high-order accurate topology optimization, useful for flows driven by natural convection. To obtain a design with a given length scale, we apply regularization techniques by filtering the material distribution. Different types of filters are applied and compared for obtaining the most realistic solution. Given the large scale of the problem, the optimization is solved with a gradient based method that relies on adjoint sensitivity analysis. The results show both the feasibility of the method, but also reveal new physical effects identified by the optimization algorithm.