Installing TensorFlow with Anaconda for your GPU

When creating an environment with Anaconda, the key is to install cuda and cudnn before TensorFlow.

But let's first set a virtual environment using Anaconda:

conda create --name my_env python=3.7

Then when it's finished, you can install the package cudatoolkit and cudnn from conda directly. But first, be sure you download the right version! TensorFlow builds are compatible with specific cuda versions. In my case, TensorFlow 2.0 is compatible with cuda 10.0 so I have to install this specific version.

You can check which version and which build is available:

conda search cudatoolkit
                                                                                                                                                                        Loading channels: done
# Name                       Version           Build  Channel
cudatoolkit                      9.0      h13b8566_0  pkgs/main
cudatoolkit                      9.2               0  pkgs/main
cudatoolkit                 10.0.130               0  pkgs/main
cudatoolkit                 10.1.168               0  pkgs/main

As we can see cuda 10.0 is available. Now let's take a loot at cudnn :

conda search cudnn

Loading channels: done
# Name                       Version           Build  Channel
cudnn                          7.0.5       cuda8.0_0  pkgs/main
cudnn                          7.1.2       cuda9.0_0  pkgs/main
cudnn                          7.1.3       cuda8.0_0  pkgs/main
cudnn                          7.2.1       cuda9.2_0  pkgs/main
cudnn                          7.3.1      cuda10.0_0  pkgs/main
cudnn                          7.3.1       cuda9.0_0  pkgs/main
cudnn                          7.3.1       cuda9.2_0  pkgs/main
cudnn                          7.6.0      cuda10.0_0  pkgs/main
cudnn                          7.6.0      cuda10.1_0  pkgs/main
cudnn                          7.6.0       cuda9.0_0  pkgs/main
cudnn                          7.6.0       cuda9.2_0  pkgs/main

Here you can see there is version 7.6.0 for cuda 10.0. So we can install those packages:

conda install cudatoolkit=10.0.130

And:

conda install cudnn=7.6.0=cuda10.0_0

Then you can install TensorFlow:

pip install --upgrade tensorflow-gpu

Finally, to make sure everything works correctly, in your conda environment, you can run:

python
Python 3.7.3 (default, Mar 27 2019, 22:11:17)
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.

Make sure the output looks similar.

Then:

>>> import tensorflow as tf
>>> tf.__version__
'2.0.0'
>>> print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU')))
2019-10-08 15:45:03.417190: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2019-10-08 15:45:03.425485: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1006] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2019-10-08 15:45:03.425762: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 0 with properties:
name: GeForce GTX 1050 Ti with Max-Q Design major: 6 minor: 1 memoryClockRate(GHz): 1.4175
pciBusID: 0000:01:00.0
2019-10-08 15:45:03.426297: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.0
2019-10-08 15:45:03.428300: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10.0
2019-10-08 15:45:03.429670: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10.0
2019-10-08 15:45:03.430822: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10.0
2019-10-08 15:45:03.433016: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10.0
2019-10-08 15:45:03.434727: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10.0
2019-10-08 15:45:03.439178: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2019-10-08 15:45:03.439337: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1006] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2019-10-08 15:45:03.439606: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1006] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2019-10-08 15:45:03.439772: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1746] Adding visible gpu devices: 0
Num GPUs Available:  1

And you can also run this small snippet from tensorflow.org to be sure:

tf.debugging.set_log_device_placement(True)

# Create some tensors
a = tf.constant([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
b = tf.constant([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]])
c = tf.matmul(a, b)

print(c)

If the code is running without any errors, your GPU is ready to work hard!