- Assume Anaconda 3 is installed under Windows 10 with Python 3.7.0.
- Tensorflow doesn't work with Python 3.7, so we will need to install a new Python 3.6 environment.
- Secondly, we will install a new 3.7 test environment to test the local pip install of a local python package.
In the Anaconda Prompt to get a list of environments:
conda info --envs or conda env list
Find the available packages named Python:
conda search python
Create a new environment for Python 3.6
conda create -n py36 python=3.6 anaconda
Activate the new environment:
conda activate py36
Check to see the Python version:
python --version
Deactivate this environment and go back to the base:
conda deactivate or simply activate the (base) environment conda activate base
Note that this py36 environment will only have the basic packages installed. If you want to install numpy, scipy etc, you will need to do so manually. For example to install numpy:
activate py36 conda install numpy deactivate
We want to create a clone of our (base) 3.7 environment so we do all our development in that new environment (e.g. so we can install local packages without polluting the base environment.
conda create --name miketest --clone base
In theory you can install tensorflow using pip
pip install tensorflow
However, to get the whole thing going (especially for GPU) there are a bunch of dependencies and I found it easier to use the anaconda package.
You can find the details at the Anaconda TensorFlow page.
I created a new conda environment called tp-gpu and installed tensorflow gpu into it as follows:
conda create -n tf-gpu tensorflow-gpu conda activate tf-gpu
This installs all the dependencies such as cudatoolkit and cudnn. The complete list of packages installed in the environment as as follows:
(tf-gpu) mikepsn@corsair-two:~/code$ conda list
# packages in environment at /home/mikepsn/anaconda3/envs/tf-gpu:
#
# Name Version Build Channel
_libgcc_mutex 0.1 main
_tflow_select 2.1.0 gpu
absl-py 0.8.1 py37_0
astor 0.8.0 py37_0
blas 1.0 mkl
c-ares 1.15.0 h7b6447c_1001
ca-certificates 2019.11.27 0
certifi 2019.11.28 py37_0
cudatoolkit 10.0.130 0
cudnn 7.6.5 cuda10.0_0
cupti 10.0.130 0
gast 0.2.2 py37_0
google-pasta 0.1.8 py_0
grpcio 1.16.1 py37hf8bcb03_1
h5py 2.9.0 py37h7918eee_0
hdf5 1.10.4 hb1b8bf9_0
intel-openmp 2019.4 243
keras-applications 1.0.8 py_0
keras-preprocessing 1.1.0 py_1
ld_impl_linux-64 2.33.1 h53a641e_7
libedit 3.1.20181209 hc058e9b_0
libffi 3.2.1 hd88cf55_4
libgcc-ng 9.1.0 hdf63c60_0
libgfortran-ng 7.3.0 hdf63c60_0
libprotobuf 3.11.2 hd408876_0
libstdcxx-ng 9.1.0 hdf63c60_0
markdown 3.1.1 py37_0
mkl 2019.4 243
mkl-service 2.3.0 py37he904b0f_0
mkl_fft 1.0.15 py37ha843d7b_0
mkl_random 1.1.0 py37hd6b4f25_0
ncurses 6.1 he6710b0_1
numpy 1.17.4 py37hc1035e2_0
numpy-base 1.17.4 py37hde5b4d6_0
openssl 1.1.1d h7b6447c_3
opt_einsum 3.1.0 py_0
pip 19.3.1 py37_0
protobuf 3.11.2 py37he6710b0_0
python 3.7.6 h0371630_1
readline 7.0 h7b6447c_5
scipy 1.3.2 py37h7c811a0_0
setuptools 44.0.0 py37_0
six 1.13.0 py37_0
sqlite 3.30.1 h7b6447c_0
tensorboard 2.0.0 pyhb38c66f_1
tensorflow 2.0.0 gpu_py37h768510d_0
tensorflow-base 2.0.0 gpu_py37h0ec5d1f_0
tensorflow-estimator 2.0.0 pyh2649769_0
tensorflow-gpu 2.0.0 h0d30ee6_0
termcolor 1.1.0 py37_1
tk 8.6.8 hbc83047_0
werkzeug 0.16.0 py_0
wheel 0.33.6 py37_0
wrapt 1.11.2 py37h7b6447c_0
xz 5.2.4 h14c3975_4
zlib 1.2.11 h7b6447c_3
First create a new environment:
conda create -n pytorch-gpu
Next install PyTorch from the pytorch conda channel:
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch
For the CPU only version:
conda install pytorch torchvision cpuonly -c pytorch
Note that TensorFlow won't see your GPU if you have the standard open source Nouveau nvidia driver installed. You need to blacklist this driver and install Nvidia's closed source binary blob driver.
Instructions can be found here and here.
There are commands in TensorFlow to show if you are using the correct driver and to see if it can see the GPU.
Details on the GPU can be found by using the nvidia-smi command
nvidia-smi
(tf-gpu) mikepsn@corsair-two:~/code$ nvidia-smi
Thu Mar 5 14:29:51 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.50 Driver Version: 430.50 CUDA Version: 10.1 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 108... Off | 00000000:01:00.0 On | N/A |
| 29% 33C P5 18W / 250W | 568MiB / 11153MiB | 1% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1230 G /usr/lib/xorg/Xorg 283MiB |
| 0 1556 G /usr/bin/kwin_x11 160MiB |
| 0 1564 G /usr/bin/krunner 2MiB |
| 0 1566 G /usr/bin/plasmashell 91MiB |
| 0 1885 G /usr/lib/firefox/firefox 2MiB |
| 0 1922 G /usr/lib/firefox/firefox 2MiB |
| 0 1980 G /usr/lib/firefox/firefox 2MiB |
| 0 2000 G /usr/lib/firefox/firefox 3MiB |
| 0 2022 G /usr/lib/firefox/firefox 7MiB |
| 0 2052 G /usr/lib/firefox/firefox 2MiB |
| 0 2078 G /usr/lib/firefox/firefox 2MiB |
| 0 2104 G /usr/lib/firefox/firefox 2MiB |
+-----------------------------------------------------------------------------+
General details on your system can be found using neofetch:
(tf-gpu) mikepsn@corsair-two:~/code$ neofetch
.-/+oossssoo+/-. mikepsn@corsair-two
`:+ssssssssssssssssss+:` -------------------
-+ssssssssssssssssssyyssss+- OS: Ubuntu 19.10 x86_64
.ossssssssssssssssssdMMMNysssso. Host: CORSAIR ONE V1
/ssssssssssshdmmNNmmyNMMMMhssssss/ Kernel: 5.3.0-40-generic
+ssssssssshmydMMMMMMMNddddyssssssss+ Uptime: 1 day, 4 hours, 44 mins
/sssssssshNMMMyhhyyyyhmNMMMNhssssssss/ Packages: 2445 (dpkg), 7 (snap)
.ssssssssdMMMNhsssssssssshNMMMdssssssss. Shell: bash 5.0.3
+sssshhhyNMMNyssssssssssssyNMMMysssssss+ Resolution: 2560x1600
ossyNMMMNyMMhsssssssssssssshmmmhssssssso DE: KDE
ossyNMMMNyMMhsssssssssssssshmmmhssssssso WM: KWin
+sssshhhyNMMNyssssssssssssyNMMMysssssss+ Theme: Breeze Dark [KDE], Breeze [GTK3]
.ssssssssdMMMNhsssssssssshNMMMdssssssss. Icons: breeze-dark [KDE], breeze [GTK3]
/sssssssshNMMMyhhyyyyhdNMMMNhssssssss/ Terminal: konsole
+sssssssssdmydMMMMMMMMddddyssssssss+ Terminal Font: Hack 11
/ssssssssssshdmNNNNmyNMMMMhssssss/ CPU: Intel i7-8700K (12) @ 4.700GHz
.ossssssssssssssssssdMMMNysssso. GPU: NVIDIA GeForce GTX 1080 Ti
-+sssssssssssssssssyyyssss+- Memory: 4446MiB / 15949MiB
`:+ssssssssssssssssss+:`
.-/+oossssoo+/-.