the runtime of a SPECFEM3D_GLOBE simulation is independent of having a 1D or 3D model. that is, running a simulation for a 1D model is as expensive as running it for a 3D model. so it doesn’t matter if you only want a portion of the Earth to be 3D.
you could only save time by turning off some of the physics (attenuation, ocean, gravity, rotation) or by limiting the simulation to a 1-chunk, regional simulation.
concerning the maximum frequency range for global simulations, Laura went down to 2s in 2008 on Ranger (High-frequency simulations of global seismic wave propagation using SPECFEM3D_GLOBE on 62K processors - IEEE Conference Publication), Seiji went down to 1.2s in 2016 on the K computer (https://journals.sagepub.com/doi/full/10.1177/1094342016632596). having a global 2Hz simulation is not achievable yet with current HPC systems, mostly due to memory limitations.
try looking into AxiSem and AxiSEM3D (GitHub - kuangdai/AxiSEM3D) which might help you with your 2Hz setup (https://academic.oup.com/gji/article/217/3/2125/5333339).