TensorFlow 报错 failed call to cuDevicePrimaryCtxRetain: CUDA_ERROR_INVALID_DEVICE

今天使用python virtualenv安装TensorFlow gpu版，一直报错

各种配置版本按官网来的https://www.tensorflow.org/

cuda8.0 cudnn5.1

错误如下

2017-06-16 13:53:53.632070: W tensorflow/stream_executor/cuda/cuda_driver.cc:523] A non-primary context 0x3f287f0 exists before initializing the StreamExecutor. We haven't verified StreamExecutor works with that.
2017-06-16 13:53:53.632367: E tensorflow/core/common_runtime/direct_session.cc:138] Internal: failed initializing StreamExecutor for CUDA device ordinal 1: Internal: failed call to cuDevicePrimaryCtxRetain: CUDA_ERROR_INVALID_DEVICE
Traceback (most recent call last):
  File "", line 1, in
  File "/home/zw/jy/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 1292, in __init__
    super(Session, self).__init__(target, graph, config=config)
  File "/home/zw/jy/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 562, in __init__
    self._session = tf_session.TF_NewDeprecatedSession(opts, status)
  File "/usr/lib/python3.5/contextlib.py", line 66, in __exit__
    next(self.gen)
  File "/home/zw/jy/lib/python3.5/site-packages/tensorflow/python/framework/errors_impl.py", line 466, in raise_exception_on_not_ok_status
    pywrap_tensorflow.TF_GetCode(status))
tensorflow.python.framework.errors_impl.InternalError: Failed to create session.

后来参考一下连接解决错误

https://stackoverflow.com/questions/37893755/tensorflow-set-cuda-visible-devices-within-jupyter

https://stackoverflow.com/questions/44375177/tensorflow-unusual-cuda-related-error

http://www.cnblogs.com/darkknightzh/p/6591923.html

原因是在创建session时没有使用我想让它用的gpu

所以需要设置

CUDA_VISIBLE_DEVICES

方法1 在python程序中

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1"

方法2

运行时 

CUDA_VISIBLE_DEVICES=1 python **.py

方法3

直接在环境变量中 export