TensorFlow Can't Detect GPU (4090) with cuDNN 8.6 and CUDA 11.8 Installed

[johntjz]

Hi everyone,

I’m currently facing an issue where TensorFlow is unable to detect my GPU and is only using the CPU. Here are the details of my setup:

• GPU: NVIDIA RTX 4090 with the latest graphics drivers installed
• CUDA Version: 11.8
• cuDNN Version: 8.6
• Operating System: [Your OS, e.g., Windows 10, Ubuntu 20.04, etc.]
• Python Version: [Your Python version, e.g., 3.8, 3.9, etc.]
• TensorFlow Version: [Your TensorFlow version, e.g., 2.10, 2.11, etc.]

I’ve made sure that the environment variables for CUDA and cuDNN are set correctly, and the necessary Python libraries are installed. Despite this, TensorFlow still does not recognize the GPU. The output from TensorFlow indicates that no GPU devices are found, and it defaults to using the CPU.

Here is what I’ve tried so far:

1. Verified that CUDA and cuDNN paths are correctly set in the environment variables.
2. Ensured that the NVIDIA drivers are up to date.
3. Reinstalled TensorFlow with GPU support.
4. Checked with nvidia-smi to confirm that the GPU is properly detected by the system.

Is there something I might be missing or any additional steps I can take to resolve this issue?

Any help or suggestions would be greatly appreciated. Thank you!

[veraj]

Hi, @johntjz

Maybe you can get GitHub - NVIDIA/cuda-samples: Samples for CUDA Developers which demonstrates features in CUDA Toolkit to run firstly to check if this issue is only specified to the Tensorflow.

Also please note this is forum for support developer tools cuda-gdb.
For TensorFlow support, please check in Deep Learning related forum. For other set up support, please check in “CUDA Setup and Installation” forum.