Mask_rcnn: GPU not in use

[ehsanpi]

The code runs only on CPU. I guess the error arises due to the this message:
" Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2".
Can you kindly explain why?

BTW, I followed all instructions and requirements both in Conda and without. I successfully ran the cuda sample with GPU full usage.

OS: Windows 10
GPU: NVIDIA 1080
Cuda: 9.0
cuDNN: 7.4.2
Python: 3.6

[4sfaloth]

" Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2".

That error message is unrelated I think. It just means that, in order to get maximum speed when running in CPU, you must build Tensorflow yourself with AVX2 enabled. They are disabled by default so that the distributed binaries work in all machines (and some do not support AVX2)

As to not using GPU, I can think of two reasons for that:
In the config.py file, is GPU_COUNT set to 1? If it is set to 0 it will not use it
Do you have tensorflow-gpu installed? If you only have tensorflow it will run on CPU

[ehsanpi]

Thanks 4sfaloth. It was great. I could install it properly.

[mymultiverse]

In the config.py file, is GPU_COUNT set to 1? If it is set to 0 it will not use it

But here it says 1 for CPU
https://github.com/matterport/Mask_RCNN/blob/master/mrcnn/config.py#L27

Correct me if I am missing something.

[qchenclaire]

The code runs only on CPU. I guess the error arises due to the this message:
" Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2".
Can you kindly explain why?

BTW, I followed all instructions and requirements both in Conda and without. I successfully ran the cuda sample with GPU full usage.

OS: Windows 10
GPU: NVIDIA 1080
Cuda: 9.0
cuDNN: 7.4.2
Python: 3.6

The code runs only on CPU. I guess the error arises due to the this message:
" Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2".
Can you kindly explain why?

BTW, I followed all instructions and requirements both in Conda and without. I successfully ran the cuda sample with GPU full usage.

OS: Windows 10
GPU: NVIDIA 1080
Cuda: 9.0
cuDNN: 7.4.2
Python: 3.6

When you run pip3 install -r requirements.txt, a cpu version of tensorflow is automatically installed. Uninstall it and reinstall a gpu version

[liuxin00738]

Can you provide some suggestions on how to install the required GPU version correctly? I am having trouble install it. I am trying to install it inside a gpu tensorflow docker.

[ouguozhen]

Can you provide some suggestions on how to install the required GPU version correctly? I am having trouble install it. I am trying to install it inside a gpu tensorflow docker.

pip uninstall tensorflow && pip install tensorflow-gpu

[ngocson1804]

" Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2".

That error message is unrelated I think. It just means that, in order to get maximum speed when running in CPU, you must build Tensorflow yourself with AVX2 enabled. They are disabled by default so that the distributed binaries work in all machines (and some do not support AVX2)

As to not using GPU, I can think of two reasons for that:
In the config.py file, is GPU_COUNT set to 1? If it is set to 0 it will not use it
Do you have tensorflow-gpu installed? If you only have tensorflow it will run on CPU

I hev the same issue. But I already install tenorflow-gpu and set GPU_COUNT = 2. it still run on CPU, not on GPU. Could you help me?