Not able to run on GPU

[tkanvar]

Hi,

I have replaced "tf.device("cpu"):" with "tf.device("/device:gpu:0"):" everywhere in the code, but it still is giving the error of "Default AvgPoolingOp only supports NHWC on device type CPU".

The code is still considering the machine as CPU instead of GPU.

Please tell the possible issues or solutions.

[latstars]

Hi,

I have replaced "tf.device("cpu"):" with "tf.device("/device:gpu:0"):" everywhere in the code, but it still is giving the error of "Default AvgPoolingOp only supports NHWC on device type CPU".

The code is still considering the machine as CPU instead of GPU.

Please tell the possible issues or solutions.

Hi, I meet the same problem with you.
I solved it with the following steps. (Maybe some of them are unnecessary.)

1. create python 2.7 environment
2. pip install tensorflow-gpu==1.4
3. change all the code of tf.device("/cpu:0") to tf.device("/gpu:0")
4. run with CUDA_VISIBLE_DEVICES=0 sh script/cifar10_micro_search.sh