tensorflow how to use multiple cpu

Question

how can I manually assign more than one device to one single op in TensorFlow (python API)?

For example, in order to assign one cpu to a graph, I would do:

with tf.device("/cpu:0"):
    # Define operation here

What if I want to assign cpu 0, 1, 2, 3 all to this same op?

It seems like I can't use a list of device names as the arg for tf.device. Any help or guidance would be appreciated. Thanks!

By default all cpus are used automatically and cpu:0 refers to all cpus on machine — Yaroslav Bulatov
– Yaroslav Bulatov, Commented Aug 5, 2017 at 0:33

darksky · Accepted Answer · 2017-08-05 00:29:29Z

1

with tf.device("/cpu:0") as cpu0, tf.device("/cpu:1") as cpu1:
    pass

answered Aug 5, 2017 at 0:29

darksky

2,12018 silver badges31 bronze badges

Sign up to request clarification or add additional context in comments.

1 Answer 1