There are several super node solutions/servers.
- huawei CloudMatrix 384
- NVIDIA DGX nvl72 / GB200
- ScaleX640
For super nodes, the network performance would be different in cpu-partition/zone/super pod. In the current node feature discovery, the topo label may be not labeled.
Is this in scope of dranet?