Serve the same model replicas on the same GPU

Ray Serve
1

Is it possible to deploy multiple replicas of the same model on the same GPU? or it’s a must to allocate a different GPU for each replica?