which cross_device_ops we need to choose for tf.distribute.MirroredStrategy?

Question

I\'m trying to train a model using tf.distribute.MirroredStrategy with two RTX 3090 cards w/o a nvlink.

In this case, which cross_device_ops will be the