You would benefit from it with some GPU offloading, this would considerably accelerate the speed of the answers. But you only need enough RAM to load the model at the bare minimum.