Using zero3 on multiple nodes is slow

I have multiple nodes, each with 8 40G A100, and I want to train a 72B model
When using zero3, the 72B model is distributed to all GPUs of all nodes. Even with nvlink, the communication delay is still very high, resulting in slow training speed, much slower than using zero3+offloading for a single node. The problem is that the more nodes there are, the slower the training speed. It is better to use only a single node
Is there a way to control zero3 to only allocate model parameters to the same node, where each node stores a complete model and only uses synchronous gradients between nodes to speed up training

The issue names no files, tests, or entry points. Start by reproducing the multi-node ZeRO-3 slowdown and reading the relevant ZeRO-3 configuration and distributed-training documentation; done would mean a supported way to keep complete model parameters within each node while synchronizing gradients between nodes.