我正在研究具有张量流的深度学习算法的GPU计算，有时会遇到奇怪的结果。 例如，当我使用经过预先训练的网络，该网络在CPU上运行良好（虽然速度很慢）以及使用以下配置的gpu：
config = tf.ConfigProto(device_count={'GPU': 0})
它工作正常但不如我预期的那么快。问题是当我使用以下配置时
config = tf.ConfigProto(allow_soft_placement=True)
变快但结果错误!!
如果您遇到此类问题，我将非常感谢您的评论。
以下是一些使用设备的方法：我搜索了它在哪里使用cpu，我发现了以下日志：
-04-19 12:05:56.359434: I tensorflow/core/common_runtime/placer.cc:884] group_deps: (NoOp)/job:localhost/replica:0/task:0/device:GPU:0
2018-04-19 12:05:56.359452: I tensorflow/core/common_runtime/placer.cc:884] keras_learning_phase/input: (Const)/job:localhost/replica:0/task:0/device:CPU:0
mbox_conf_final/Exp: (Exp): /job:localhost/replica:0/task:0/device:GPU:0
（2）
2018-04-19 12:05:56.351642: I tensorflow/core/common_runtime/placer.cc:884] Assign_42: (Assign)/job:localhost/replica:0/task:0/device:GPU:0
keras_learning_phase: (PlaceholderWithDefault): /job:localhost/replica:0/task:0/device:CPU:0
non_max_suppression/NonMaxSuppressionV2: (NonMaxSuppressionV2): /job:localhost/replica:0/task:0/device:CPU:0
pool6_mbox_priorbox/Variable: (VariableV2): /job:localhost/replica:0/task:0/device:GPU:0
（3）
2018-04-19 12:05:56.348320: I tensorflow/core/common_runtime/placer.cc:884] keras_learning_phase: (PlaceholderWithDefault)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:56.348382: I tensorflow/core/common_runtime/placer.cc:884] non_max_suppression/NonMaxSuppressionV2: (NonMaxSuppressionV2)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:56.348446: I tensorflow/core/common_runtime/placer.cc:884] pool6_mbox_priorbox/Variable: (VariableV2)/job:localhost/replica:0/task:0/device:GPU:0
（4）
2018-04-19 12:05:44.754020: I tensorflow/core/common_runtime/placer.cc:884] non_max_suppression/iou_threshold: (Const)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754038: I tensorflow/core/common_runtime/placer.cc:884] Placeholder_68: (Placeholder)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754056: I tensorflow/core/common_runtime/placer.cc:884] Placeholder_67: (Placeholder)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754075: I tensorflow/core/common_runtime/placer.cc:884] Placeholder_66: (Placeholder)/job:localhost/replica:0/task:0/device:GPU:0
（5）
2018-04-19 12:05:44.754001: I tensorflow/core/common_runtime/placer.cc:884] non_max_suppression/NonMaxSuppressionV2/max_output_size: (Const)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754020: I tensorflow/core/common_runtime/placer.cc:884] non_max_suppression/iou_threshold: (Const)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754038: I tensorflow/core/common_runtime/placer.cc:884] Placeholder_68: (Placeholder)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754056: I tensorflow/core/common_runtime/placer.cc:884] Placeholder_67: (Placeholder)/job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.754075: I tensorflow/core/common_runtime/placer.cc:884] Placeholder_66: (Placeholder)/job:localhost/replica:0/task:0/device:GPU:0
（6）
group_deps: (NoOp): /job:localhost/replica:0/task:0/device:GPU:0
non_max_suppression/NonMaxSuppressionV2/max_output_size: (Const): /job:localhost/replica:0/task:0/device:CPU:0
non_max_suppression/iou_threshold: (Const): /job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.753194: I tensorflow/core/common_runtime/placer.cc:884] conv8_2_mbox_priorbox/Tile/multiples: (Pack)/job:localhost/replica:0/task:0/device:GPU:0
Placeholder_68: (Placeholder): /job:localhost/replica:0/task:0/device:CPU:0
2018-04-19 12:05:44.753212: I tensorflow/core/common_runtime/placer.cc:884] conv8_2_mbox_priorbox/Tile: (Tile)/job:localhost/replica:0/task:0/device:GPU:0
Placeholder_67: (Placeholder): /job:localhost/replica:0/task:0/device:CPU:0
Placeholder_66: (Placeholder): /job:localhost/replica:0/task:0/device:GPU:0