Based on this guide1
Installing packages
Important note: You should start by installing Bumblebee and making sure it works: (Arch Linux Wiki Bumblebee
My package versions:
nvidia 396.24-7
python-numpy 1.14.3-1
swig 3.0.12-1
cudnn 7.1.4-1
cuda 9.0.176-4
python 3.6.5-3
python-virtualenv 15.1.0-3
pacman -S nvidia python-numpy swig cudnn cuda python3 python-virtualenv
Then reboot. This should install all necessary packages.
In my case I also had to downgrade CUDA to version 9.x. I used this AUR Package.
sudo downgrade cuda
(and select the highest version 9.0.x)
Virtualenv
My PIP packages:
absl-py==0.2.2
astor==0.6.2
bleach==1.5.0
gast==0.2.0
grpcio==1.12.1
html5lib==0.9999999
Markdown==2.6.11
numpy==1.14.3
protobuf==3.5.2.post1
six==1.11.0
tensorboard==1.8.0
tensorflow-gpu==1.8.0
termcolor==1.1.0
Werkzeug==0.14.1
Next create a new Python3 virtualenv.
virtualenv -p python3 tensorflow-env
source tensorflow/bin/activate
We’ll need to update PIP and install the latest tensorflow-gpu package
pip install --upgrade pip
pip install --upgrade tensorflow-gpu
Validating our installation
Based on this guide2
To validate our installation we’ll use the following Python file
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))
I saved it to test.py, and ran it using the command:
optirun python test.py
(optirun is required to turn on the GPU)
The output should be something similar to this:
2018-06-05 09:44:13.468281: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2018-06-05 09:44:13.545180: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-06-05 09:44:13.545552: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1356] Found device 0 with properties:
name: GeForce GTX 1050 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.62
pciBusID: 0000:01:00.0
totalMemory: 3.95GiB freeMemory: 3.89GiB
2018-06-05 09:44:13.545566: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-06-05 09:44:13.733866: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-06-05 09:44:13.733913: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929] 0
2018-06-05 09:44:13.733919: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0: N
2018-06-05 09:44:13.734069: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3623 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1050 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1)
b'Hello, TensorFlow!'
(tensorflow)
If you run it without optirun it will look like this (does not detect GPU):
python test.py
2018-06-05 09:11:15.197152: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2018-06-05 09:11:15.375056: E tensorflow/stream_executor/cuda/cuda_driver.cc:406] failed call to cuInit: CUDA_ERROR_UNKNOWN
2018-06-05 09:11:15.375095: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:145] kernel driver does not appear to be running on this host (saber): /proc/driver/nvidia/version does not exist
b'Hello, TensorFlow!'
(tensorflow)