The current implementation has a workaround to create one ResourceSlice per NUMA node if the machine has more than 128 CPUs. This is a obvious limitation and needs a better solution.

/kind feature