Ubuntu 18.04下TensorFlow 2.2 GPU安装失败求助及规范步骤需求

Hey there, I’ve been through this exact issue with mismatched CUDA libraries and GPU detection in TensorFlow 2.2 on Ubuntu 18.04. Let’s walk through a strict, version-aligned setup from a fresh server—this will eliminate those libcublas.so errors and ensure TensorFlow can talk to your GPU properly.

1. Update System & Install Prerequisites #

First, get your server up to date and install essential build tools:

sudo apt update && sudo apt upgrade -y
sudo apt install -y build-essential cmake git unzip zip zlib1g-dev

2. Install NVIDIA GPU Driver (Compatible with CUDA 10.1) #

TensorFlow 2.2 relies on CUDA 10.1, which requires NVIDIA driver version 418.x. We’ll use the official PPA for a hassle-free install:

# Add NVIDIA graphics drivers repo
sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt update

# Install the required driver version
sudo apt install -y nvidia-driver-418

# Reboot to apply the driver (critical step!)
sudo reboot

After rebooting, verify the driver works with:

nvidia-smi

You should see your GPU details and driver version (418.x) listed.

3. Install CUDA Toolkit 10.1 (Exact Version for TF 2.2) #

Do NOT install the latest CUDA version—TF 2.2 only supports 10.1. Follow these steps:

# Download and set up CUDA 10.1 repo
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget http://developer.download.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda-repo-ubuntu1804-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-10-1-local-10.1.243-418.87.00/7fa2af80.pub
sudo apt update

# Install CUDA 10.1 (skip the driver since we already installed it)
sudo apt install -y cuda-toolkit-10.1

Set CUDA Environment Variables #

Add these to your shell profile to ensure system-wide access to CUDA libraries:

# Append to ~/.bashrc
echo 'export PATH=/usr/local/cuda-10.1/bin${PATH:+:${PATH}}' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=/usr/local/cuda-10.1/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}' >> ~/.bashrc

# Apply changes immediately
source ~/.bashrc

Verify CUDA installation with:

nvcc --version

You should see release 10.1 in the output.

4. Install cuDNN 7.6.5 (Compatible with CUDA 10.1 & TF 2.2) #

cuDNN is required for GPU-accelerated TensorFlow operations. We need version 7.6.5 for this setup:

# Download cuDNN packages (matches CUDA 10.1)
wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/libcudnn7_7.6.5.32-1+cuda10.1_amd64.deb
wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/libcudnn7-dev_7.6.5.32-1+cuda10.1_amd64.deb

# Install the packages
sudo dpkg -i libcudnn7_7.6.5.32-1+cuda10.1_amd64.deb
sudo dpkg -i libcudnn7-dev_7.6.5.32-1+cuda10.1_amd64.deb

# Update system library cache
sudo ldconfig

5. Set Up Python Virtual Environment (Avoid Dependency Conflicts) #

Using a virtual environment prevents conflicts with system Python packages:

# Install Python 3.7 (TF 2.2 works best with Python 3.6-3.8)
sudo apt install -y python3.7 python3.7-dev python3-pip

# Install virtualenv
sudo pip3 install virtualenv

# Create a dedicated environment for TF 2.2 GPU
virtualenv -p python3.7 tf22-gpu-env

# Activate the environment (you'll need to run this every time you work with TF)
source tf22-gpu-env/bin/activate

6. Install TensorFlow 2.2 GPU #

Now install the exact GPU version of TensorFlow:

pip install tensorflow-gpu==2.2.0

7. Verify GPU Support #

Run this quick test to confirm TensorFlow can detect and use your GPU:

python -c "
import tensorflow as tf
print('TensorFlow Version:', tf.__version__)
print('GPU Available:', tf.test.is_gpu_available())
print('Detected GPUs:', tf.config.list_physical_devices('GPU'))
"

If everything works, you’ll see GPU Available: True and a list of your GPU devices.

内容的提问来源于stack exchange，提问作者Basj