TensorFlow model training can spend most of its time in matrix operations that an NVIDIA GPU handles faster than a CPU. A project uses that accelerator only when the active Python environment can load both the TensorFlow package and the host GPU driver, so the environment needs a GPU-aware install and a device check before training starts.

Current Linux x86_64 and WSL2 installs use the pip package extra tensorflow[and-cuda] inside the environment that will run the model code. That package installs TensorFlow plus matching CUDA and cuDNN user-space libraries, while the host still provides the NVIDIA kernel driver that nvidia-smi reports.

Native Windows GPU support stopped after TensorFlow 2.10, and current macOS packages run on the CPU path. If the host driver cannot see a GPU first, reinstalling TensorFlow will not make GPU devices appear inside tf.config.list_physical_devices('GPU').

Related: How to build a TensorFlow GPU environment
Related: How to fix TensorFlow not detecting a GPU
Related: How to run distributed GPU training in TensorFlow

Steps to enable TensorFlow GPU acceleration:

  1. Activate the Python environment that should run TensorFlow on the GPU. 
    $ source ~/venvs/tf-gpu/bin/activate
(tf-gpu) $

    Use conda activate <name> instead when the project environment is managed by Conda.
    Related: How to create a virtual environment for TensorFlow
    Related: How to create a Conda environment for TensorFlow

  2. Confirm that the host NVIDIA driver can see the GPU. 
    (tf-gpu) $ nvidia-smi
Mon Jun 29 04:20:31 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 575.64.03              Driver Version: 575.64.03      CUDA Version: 12.9     |
|-----------------------------------------+------------------------+----------------------|
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
|  0   NVIDIA RTX 4080 SUPER         Off  | 00000000:01:00.0  Off |                  Off |
##### snipped #####

    If nvidia-smi fails or prints no GPU, repair the Linux driver or WSL2 GPU passthrough before changing TensorFlow packages. The pip package cannot replace a missing host driver.

  3. Upgrade pip inside the active environment. 
    (tf-gpu) $ python -m pip install --upgrade pip
Requirement already satisfied: pip in /home/user/venvs/tf-gpu/lib/python3.12/site-packages (24.0)
Collecting pip
##### snipped #####
Successfully installed pip-26.0.1
  4. Install or upgrade the current TensorFlow GPU package. 
    (tf-gpu) $ python -m pip install --upgrade "tensorflow[and-cuda]"
Collecting tensorflow[and-cuda]
  Downloading tensorflow-2.21.0-cp312-cp312-manylinux_2_27_x86_64.whl.metadata
Collecting nvidia-cudnn-cu12&lt;10.0,&gt;=9.3.0.75
##### snipped #####
Successfully installed keras-3.14.0 nvidia-cudnn-cu12-9.21.0.82 tensorflow-2.21.0

    The and-cuda extra keeps CUDA and cuDNN user-space libraries inside the active Python environment.
    Related: How to install TensorFlow with pip

  5. Confirm that the installed TensorFlow package imports from the active environment. 
    (tf-gpu) $ python -c "import tensorflow as tf; print(tf.__version__)"
2.21.0

    Related: How to check the TensorFlow version

  6. List the physical GPU devices that TensorFlow can use. 
    (tf-gpu) $ python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

    If the output is [], TensorFlow imported but did not find a usable GPU runtime path.

  7. Move into the installed TensorFlow package directory if the GPU list is empty. 
    (tf-gpu) $ pushd $(dirname $(python -c 'print(__import__("tensorflow").__file__)'))
~/venvs/tf-gpu/lib/python3.12/site-packages/tensorflow ~/project
  8. Create the shared-library symlinks if the GPU list is empty. 
    (tf-gpu) $ ln -svf ../nvidia/*/lib/*.so* .
'./libcublas.so.12' -> '../nvidia/cublas/lib/libcublas.so.12'
'./libcudart.so.12' -> '../nvidia/cuda_runtime/lib/libcudart.so.12'
##### snipped #####

    This repair follows TensorFlow's pip install notes for virtual environments where packaged NVIDIA libraries are installed but not discovered automatically.

  9. Return to the project directory after creating the library symlinks. 
    (tf-gpu) $ popd
~/project
  10. Create the ptxas symlink if the GPU list is empty. 
    (tf-gpu) $ ln -sf $(find $(dirname $(dirname $(python -c "import nvidia.cuda_nvcc; print(nvidia.cuda_nvcc.__file__)"))/*/bin/) -name ptxas -print -quit) $VIRTUAL_ENV/bin/ptxas

    Related: How to fix TensorFlow not detecting a GPU

  11. Run the GPU device check again after any symlink repair. 
    (tf-gpu) $ python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
  12. Run a small tensor operation and confirm it is placed on the GPU. 
    (tf-gpu) $ python -c "import tensorflow as tf; x=tf.ones((512,512)); print(tf.matmul(x,x).device)"
/job:localhost/replica:0/task:0/device:GPU:0

    After this command reports GPU:0, the environment is ready for GPU training options such as distribution strategies.
    Related: How to run distributed GPU training in TensorFlow