参数里没有指定gpu id，分布式时默认在每一个worker里都用第0个GPU吗

请问：
run_loop里只指定了GPU的动态增长策略，没有指定gpu id, tensorflow默认使用第0片GPU，所以
Euler也默认每个worker使用第0片GPU吗？
ps server的参数都是存在cpu的，可以存在gpu吗，如果存在gpu里是不是会快一些呢？

感谢解答

Start by reading the run_loop GPU configuration and the distributed worker setup to determine how GPU IDs are selected. Then inspect the ps server parameters and any related TensorFlow device-placement documentation. Done means documenting whether workers default to GPU 0 and whether ps servers can run on GPUs, including the expected configuration.

Written by the indexing model from the issue text.