Tensorflow并行计算中不确定性的根源是什么？

Question

要在Keras / Tensorflow中获得确定性结果，（除其他外）有必要“停用并行性”，如Keras doc中所述：

# Force TensorFlow to use single thread.
# Multiple threads are a potential source of
# non-reproducible results.
# For further details, see: https://stackoverflow.com/questions/42022950/which-seeds-have-to-be-set-where-to-realize-100-reproducibility-of-training-res

session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)

This thread还指出（在评论中）：

You also need to remove parallelism from your computation because that is often non-deterministic

是什么使Tensorflow并行计算具有不确定性？