Kubernetes with Kubeflow vs. Slurm: Scalability

Which one provides better scalability. Say I want to implement a 15k+ BM nodes HPC cluster... what would you go for?