time is 1875.640280

[bigsea00001]

I am wondering if the train screen changes every 1875 seconds and this speed is correct.
Can not use this program with CUDA?

[2/150][50] Loss_DA: 0.205057 Loss_DB: 0.193327 Loss_GA: 0.415360 Loss_GB: 0.413623 time: 1875.640280
[4/150][100] Loss_DA: 0.200707 Loss_DB: 0.211183 Loss_GA: 0.295002 Loss_GB: 0.341839 time: 3606.393274

[shaoanlu]

Please check if tensorflow is using gpu. Also,

from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())

this can list the available devices on your machine.

[bigsea00001]

Thank you for answer.
As you can see, I have device_type: "CPU" as shown below. I am using Cuda 8.0, gtx 1060 6GB, and I will look for a solution.

[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 9750307169120904566
]

[shaoanlu]

You're welcome. Good luck and happy swapping.

[BillyHalley]

@bigsea00001 Did you manage to resolve it?

[Yare-zz]

I had this issue, which is now resolved. For future reference:

I first installed tensorflow-gpu 1.5, but didn't test it. I had CUDA 8 installed (with cuDNN 5.1), but tensorflow 1.5 expects CUDA 9 (with cuDNN 7). Normally this would have been obvious the first time I tried to use the notebooks and tensorflow exploded with a missing DLL error. However, the failure was masked as a result of running this command in the notebook:

pip install keras_vggface

This package installs the CPU-based tensorflow as a dependency. And so when I ran the training notebooks everything silently "worked" as it fell back on CPU tensorflow. I resolved the issue by doing:

pip uninstall tensorflow

And then trying:

python
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())

This finally raised an error that I was missing CUDA 9 DLLs. I installed the CUDA 9 stack, relisted my local devices, and the GPU finally showed up. Training times improved drastically.

@shaoanlu , you might consider adding some of these notes to your readme. Many of your users are probably unaware that they are running in CPU. In my notebook I also threw in the list_local_devices call right after importing tensorflow so I can verify immediately if I'm working with GPU or not.