How does the cpu and gpu work together in deepmd modules ?

[12jscvb]

In the deepmd module of dpgen workflow , the resources parameter is set as follows in machine.json file:

machine.json

After submitting the running command, enter the' top' command in the total terminal. The following information is displayed:

I have set 64 cores for 1 node in machine file, why the result shows that the cpu usage is very low, while the gpu usage reaches 80%. How does the cpu and gpu of deepmd module work together? The server is configured as a single machine with two numa nodes and 64 cores for each node. Thank you for help