Tensorflow问题汇总

问题：

Cannot assign a device for operation 'MatMul': Operation was explicitly assigned to /device:GPU:1 but available devices are [ /job:localhost/replica:0/task:0/device:CPU:0, /job:localhost/replica:0/task:0/device:GPU:0 ]. Make sure the device specification refers to a valid device.

解决方法：

方法一. 卸载tensorflow，安装tensorflow-gpu

pip uninstall tensorflow

pip install tensorflow-gpu

如果还是出问题请尝试：

方法二. 将乘法操作移到gpu以外执行

示例：

# old
with tf.Session() as sess:
    matrix1 = tf.constant([[3., 3.]])
    print("matrix1:",matrix1)
    matrix2 = tf.constant([[2.],[2.]])
    print("matrix2:",matrix2)    
    with tf.device("/gpu:1"):
        # 将乘法运算移到外面执行
        product = tf.matmul(matrix1, matrix2)

        result = sess.run(product)
        print("result:",result)

# new
with tf.Session() as sess:
    matrix1 = tf.constant([[3., 3.]])
    print("matrix1:",matrix1)
    matrix2 = tf.constant([[2.],[2.]])
    print("matrix2:",matrix2)
    product = tf.matmul(matrix1, matrix2)

    with tf.device("/gpu:1"):
        result = sess.run(product)
        print("result:",result)

正确结果：

C:Usersin>python d:/PycharmProjects/TFLearn/Unit1/03.py
2018-01-11 22:43:03.490996: I C:	f_jenkinshomeworkspace
el-winMwindows-gpuPY36	ensorflowcoreplatformcpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX
2018-01-11 22:43:04.077880: I C:	f_jenkinshomeworkspace
el-winMwindows-gpuPY36	ensorflowcorecommon_runtimegpugpu_device.cc:1030] Found device 0 with properties:
name: GeForce GT 740M major: 3 minor: 5 memoryClockRate(GHz): 1.0325
pciBusID: 0000:01:00.0
totalMemory: 1.00GiB freeMemory: 836.07MiB
2018-01-11 22:43:04.078060: I C:	f_jenkinshomeworkspace
el-winMwindows-gpuPY36	ensorflowcorecommon_runtimegpugpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce GT 740M, pci bus id: 0000:01:00.0, compute capability: 3.5)
matrix1: Tensor("Const:0", shape=(1, 2), dtype=float32)
matrix2: Tensor("Const_1:0", shape=(2, 1), dtype=float32)
result: [[ 12.]]

问题：

Could not find 'cudnn64_6.dll'. TensorFlow requires that this DLL be installed in a directory that is named in your %PATH% environment variable. Note that installing cuDNN is a separate step from installing CUDA, and this DLL is often found in a different directory from the CUDA DLLs. You may install the necessary DLL by downloading cuDNN 6 from this URL: https://developer.nvidia.com/cudnn

解决方法：

1. 安装cudn: 

https://developer.nvidia.com/rdp/cudnn-download

https://developer.nvidia.com/cuda-zone

nvidia官网经常更新，请下载对应的版本，cudn和cudnn dll 版本号不一样，比如tensorflow-gpu 1.4对应cndn8，同时需要下载cudnn64_6.dll for cudn8。

2. 下载cudnn64_6，解压到指定cudn目录 C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0

pan.baidu.com/s/1o8mc4Y  windows

pan.baidu.com/s/1hs23Hr  linux