Skip to content

Latest commit

 

History

History
168 lines (126 loc) · 6.94 KB

install.md

File metadata and controls

168 lines (126 loc) · 6.94 KB
Raw

Installation of tensorflow

date: 2023 Nov 12

This installation guide is specific to the use case as in the pre-requisites. I'm not sure if it is similar for Mac/Linux native OSes or Other WSL2 Distros. Let me know either way if it does help.

Also, welcome to edit and make corrections directly (via PRs) or let me know via Slack.

pre-requisites

  • wsl2 on Windows 10 19044 or higher
  • Nvidia Pascal GPU, current driver 546.01
  • Ubuntu 22.04 LTS
  • micromamba environment manager
  • bash/zsh shell
  • cuda_12.3.0_545.84_windows.exe and cudnn-windows-x86_64-8.9.6.50_cuda12-archive.zip
  • Nvidia CUDA Toolkit 12.3 on WSL2
  • Tensorflow v2.14 with cuda

starting over or pre-install

windows host

  1. control panel --> Uninstall Cuda v??.?
  2. verify Nvidia driver is >= v450.xx (currently v546.01 as of @2023 Nov 12)
  3. download cuda executable for Win10
    • install following screen prompts, but select Custom Install, shorten the Path if desired. I've shorten from the defaults to just cuda
    • disable nsight, visual studio, GPU driver, GeForce Experience, HD Audio driver (if installed separately)
    • add to PATH manually or verify these are present; first 2 are mandatory, 3rd might be optional (I'm not sure):
      • C:\Program Files\cuda\v12.3\bin
      • C:\Program Files\cuda\v12.3\libnvvp
      • D:\Program Files\cuda\v12.3\extras\CUPTI\lib64
  4. cudnn-windows-x86_64-8.9.6.zip, download cudnn archive for other versions
    • unzip and drag contents into the C:\Program Files\cuda\v12.3\ where the matching bin, include and lib are located. Let the contents from cudnn*.zip replace the folders in the CUDA folder

wsl2 Ubuntu 22.04

source --> go to Ubuntu and Debian section

In the wsl2 Ubuntu terminal, clean up previous installs especially if on a different version

sudo apt-key del 7fa2af80
sudo apt-get --purge remove "*cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" \
 "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
sudo apt-get --purge remove "*nvidia*" "libxnvctrl*"
sudo apt-get autoremove
rm -rf "*.deb" "*.tar" "*.gz"

source: go here for exact version numbers, other than 12.3 @2023 Nov 12

wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-wsl-ubuntu.pin
sudo mv cuda-wsl-ubuntu.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/12.3.0/local_installers/cuda-repo-wsl-ubuntu-12-3-local_12.3.0-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-12-3-local_12.3.0-1_amd64.deb
sudo cp /var/cuda-repo-wsl-ubuntu-12-3-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-3

post Install

export PATH=/usr/local/cuda-12.3/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-12.3/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

create new environment

micromamba create -n py310 python=3.10

activate environment

micromamba activate py310

install tensorflow

pip install 'tensorflow[and-cuda]'
or
python3 -m pip install 'tensorflow[and-cuda]'

Outcome:

in virtual environment

  • tensorflow-2.14
  • tensorboard<2.15,>=2.14
  • keras<2.15,>=2.14.0

in wsl2

  • nvidia-cuda-runtime-cu12==11.8.89
  • nvidia-cudnn-cu12==8.7.0.84

verify install

  1. Run test_tf.ipynb to check if GPU recognized, or

  2. Run the notebook.ipynb cells, and use nvidia-smi while it's running. Ensure the correct versions are reported in nvidia-smi output.

    • Driver version
    • CUDA version
    • processes list your python version
    • if you repeatedly run it with the up-arrow key you should be able to see the values of FAN%, Temp, Wattage, Memory, and Util change each time
nvidia-smi                                            
Sun Nov 12 22:59:54 2023       
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 545.29.01              Driver Version: 546.01       CUDA Version: 12.3     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce GTX 970         On  | 00000000:01:00.0  On |                  N/A |
| 40%   61C    P2              93W / 200W |   3872MiB /  4096MiB |     54%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|    0   N/A  N/A        23      G   /Xwayland                                 N/A      |
|    0   N/A  N/A      2490      C   /python3.10                               N/A      |
+---------------------------------------------------------------------------------------+

FAQs

Issue: /sbin/ldconfig.real: /usr/lib/wsl/lib/libcuda.so.1 is not a symbolic link

Solution: paste this ldconfig = false entry into [automount] section in your %USERPROFILE%/.wslconfig file. Open the file in vs-code or text editor like notepad++. Mine's at "C:\Users\Ella\.wslconfig". I had a previous entry to cap my RAM usage in WSL2. source

[wsl2] 
memory=12GB

[automount]
ldconfig = false

|=============================================================================|

Issue: warnings and info messages

Solution: surpress with source

import os  
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

|=============================================================================|